ghcup-hs/3rdparty/libarchive/c/archive_entry_link_resolver.c

448 lines
12 KiB
C

/*-
* Copyright (c) 2003-2007 Tim Kientzle
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR(S) ``AS IS'' AND ANY EXPRESS OR
* IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
* OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
* IN NO EVENT SHALL THE AUTHOR(S) BE LIABLE FOR ANY DIRECT, INDIRECT,
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
* THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#include "archive_platform.h"
__FBSDID("$FreeBSD: head/lib/libarchive/archive_entry_link_resolver.c 201100 2009-12-28 03:05:31Z kientzle $");
#ifdef HAVE_SYS_STAT_H
#include <sys/stat.h>
#endif
#ifdef HAVE_ERRNO_H
#include <errno.h>
#endif
#include <stdio.h>
#ifdef HAVE_STDLIB_H
#include <stdlib.h>
#endif
#ifdef HAVE_STRING_H
#include <string.h>
#endif
#include "archive.h"
#include "archive_entry.h"
/*
* This is mostly a pretty straightforward hash table implementation.
* The only interesting bit is the different strategies used to
* match up links. These strategies match those used by various
* archiving formats:
* tar - content stored with first link, remainder refer back to it.
* This requires us to match each subsequent link up with the
* first appearance.
* cpio - Old cpio just stored body with each link, match-ups were
* implicit. This is trivial.
* new cpio - New cpio only stores body with last link, match-ups
* are implicit. This is actually quite tricky; see the notes
* below.
*/
/* Users pass us a format code, we translate that into a strategy here. */
#define ARCHIVE_ENTRY_LINKIFY_LIKE_TAR 0
#define ARCHIVE_ENTRY_LINKIFY_LIKE_MTREE 1
#define ARCHIVE_ENTRY_LINKIFY_LIKE_OLD_CPIO 2
#define ARCHIVE_ENTRY_LINKIFY_LIKE_NEW_CPIO 3
/* Initial size of link cache. */
#define links_cache_initial_size 1024
struct links_entry {
struct links_entry *next;
struct links_entry *previous;
struct archive_entry *canonical;
struct archive_entry *entry;
size_t hash;
unsigned int links; /* # links not yet seen */
};
struct archive_entry_linkresolver {
struct links_entry **buckets;
struct links_entry *spare;
unsigned long number_entries;
size_t number_buckets;
int strategy;
};
#define NEXT_ENTRY_DEFERRED 1
#define NEXT_ENTRY_PARTIAL 2
#define NEXT_ENTRY_ALL (NEXT_ENTRY_DEFERRED | NEXT_ENTRY_PARTIAL)
static struct links_entry *find_entry(struct archive_entry_linkresolver *,
struct archive_entry *);
static void grow_hash(struct archive_entry_linkresolver *);
static struct links_entry *insert_entry(struct archive_entry_linkresolver *,
struct archive_entry *);
static struct links_entry *next_entry(struct archive_entry_linkresolver *,
int);
struct archive_entry_linkresolver *
archive_entry_linkresolver_new(void)
{
struct archive_entry_linkresolver *res;
/* Check for positive power-of-two */
if (links_cache_initial_size == 0 ||
(links_cache_initial_size & (links_cache_initial_size - 1)) != 0)
return (NULL);
res = calloc(1, sizeof(struct archive_entry_linkresolver));
if (res == NULL)
return (NULL);
res->number_buckets = links_cache_initial_size;
res->buckets = calloc(res->number_buckets, sizeof(res->buckets[0]));
if (res->buckets == NULL) {
free(res);
return (NULL);
}
return (res);
}
void
archive_entry_linkresolver_set_strategy(struct archive_entry_linkresolver *res,
int fmt)
{
int fmtbase = fmt & ARCHIVE_FORMAT_BASE_MASK;
switch (fmtbase) {
case ARCHIVE_FORMAT_7ZIP:
case ARCHIVE_FORMAT_AR:
case ARCHIVE_FORMAT_ZIP:
res->strategy = ARCHIVE_ENTRY_LINKIFY_LIKE_OLD_CPIO;
break;
case ARCHIVE_FORMAT_CPIO:
switch (fmt) {
case ARCHIVE_FORMAT_CPIO_SVR4_NOCRC:
case ARCHIVE_FORMAT_CPIO_SVR4_CRC:
res->strategy = ARCHIVE_ENTRY_LINKIFY_LIKE_NEW_CPIO;
break;
default:
res->strategy = ARCHIVE_ENTRY_LINKIFY_LIKE_OLD_CPIO;
break;
}
break;
case ARCHIVE_FORMAT_MTREE:
res->strategy = ARCHIVE_ENTRY_LINKIFY_LIKE_MTREE;
break;
case ARCHIVE_FORMAT_ISO9660:
case ARCHIVE_FORMAT_SHAR:
case ARCHIVE_FORMAT_TAR:
case ARCHIVE_FORMAT_XAR:
res->strategy = ARCHIVE_ENTRY_LINKIFY_LIKE_TAR;
break;
default:
res->strategy = ARCHIVE_ENTRY_LINKIFY_LIKE_OLD_CPIO;
break;
}
}
void
archive_entry_linkresolver_free(struct archive_entry_linkresolver *res)
{
struct links_entry *le;
if (res == NULL)
return;
while ((le = next_entry(res, NEXT_ENTRY_ALL)) != NULL)
archive_entry_free(le->entry);
free(res->buckets);
free(res);
}
void
archive_entry_linkify(struct archive_entry_linkresolver *res,
struct archive_entry **e, struct archive_entry **f)
{
struct links_entry *le;
struct archive_entry *t;
*f = NULL; /* Default: Don't return a second entry. */
if (*e == NULL) {
le = next_entry(res, NEXT_ENTRY_DEFERRED);
if (le != NULL) {
*e = le->entry;
le->entry = NULL;
}
return;
}
/* If it has only one link, then we're done. */
if (archive_entry_nlink(*e) == 1)
return;
/* Directories, devices never have hardlinks. */
if (archive_entry_filetype(*e) == AE_IFDIR
|| archive_entry_filetype(*e) == AE_IFBLK
|| archive_entry_filetype(*e) == AE_IFCHR)
return;
switch (res->strategy) {
case ARCHIVE_ENTRY_LINKIFY_LIKE_TAR:
le = find_entry(res, *e);
if (le != NULL) {
archive_entry_unset_size(*e);
archive_entry_copy_hardlink(*e,
archive_entry_pathname(le->canonical));
} else
insert_entry(res, *e);
return;
case ARCHIVE_ENTRY_LINKIFY_LIKE_MTREE:
le = find_entry(res, *e);
if (le != NULL) {
archive_entry_copy_hardlink(*e,
archive_entry_pathname(le->canonical));
} else
insert_entry(res, *e);
return;
case ARCHIVE_ENTRY_LINKIFY_LIKE_OLD_CPIO:
/* This one is trivial. */
return;
case ARCHIVE_ENTRY_LINKIFY_LIKE_NEW_CPIO:
le = find_entry(res, *e);
if (le != NULL) {
/*
* Put the new entry in le, return the
* old entry from le.
*/
t = *e;
*e = le->entry;
le->entry = t;
/* Make the old entry into a hardlink. */
archive_entry_unset_size(*e);
archive_entry_copy_hardlink(*e,
archive_entry_pathname(le->canonical));
/* If we ran out of links, return the
* final entry as well. */
if (le->links == 0) {
*f = le->entry;
le->entry = NULL;
}
} else {
/*
* If we haven't seen it, tuck it away
* for future use.
*/
le = insert_entry(res, *e);
if (le == NULL)
/* XXX We should return an error code XXX */
return;
le->entry = *e;
*e = NULL;
}
return;
default:
break;
}
return;
}
static struct links_entry *
find_entry(struct archive_entry_linkresolver *res,
struct archive_entry *entry)
{
struct links_entry *le;
size_t hash, bucket;
dev_t dev;
int64_t ino;
/* Free a held entry. */
if (res->spare != NULL) {
archive_entry_free(res->spare->canonical);
archive_entry_free(res->spare->entry);
free(res->spare);
res->spare = NULL;
}
dev = archive_entry_dev(entry);
ino = archive_entry_ino64(entry);
hash = (size_t)(dev ^ ino);
/* Try to locate this entry in the links cache. */
bucket = hash & (res->number_buckets - 1);
for (le = res->buckets[bucket]; le != NULL; le = le->next) {
if (le->hash == hash
&& dev == archive_entry_dev(le->canonical)
&& ino == archive_entry_ino64(le->canonical)) {
/*
* Decrement link count each time and release
* the entry if it hits zero. This saves
* memory and is necessary for detecting
* missed links.
*/
--le->links;
if (le->links > 0)
return (le);
/* Remove it from this hash bucket. */
if (le->previous != NULL)
le->previous->next = le->next;
if (le->next != NULL)
le->next->previous = le->previous;
if (res->buckets[bucket] == le)
res->buckets[bucket] = le->next;
res->number_entries--;
/* Defer freeing this entry. */
res->spare = le;
return (le);
}
}
return (NULL);
}
static struct links_entry *
next_entry(struct archive_entry_linkresolver *res, int mode)
{
struct links_entry *le;
size_t bucket;
/* Free a held entry. */
if (res->spare != NULL) {
archive_entry_free(res->spare->canonical);
archive_entry_free(res->spare->entry);
free(res->spare);
res->spare = NULL;
}
/* Look for next non-empty bucket in the links cache. */
for (bucket = 0; bucket < res->number_buckets; bucket++) {
for (le = res->buckets[bucket]; le != NULL; le = le->next) {
if (le->entry != NULL &&
(mode & NEXT_ENTRY_DEFERRED) == 0)
continue;
if (le->entry == NULL &&
(mode & NEXT_ENTRY_PARTIAL) == 0)
continue;
/* Remove it from this hash bucket. */
if (le->next != NULL)
le->next->previous = le->previous;
if (le->previous != NULL)
le->previous->next = le->next;
else
res->buckets[bucket] = le->next;
res->number_entries--;
/* Defer freeing this entry. */
res->spare = le;
return (le);
}
}
return (NULL);
}
static struct links_entry *
insert_entry(struct archive_entry_linkresolver *res,
struct archive_entry *entry)
{
struct links_entry *le;
size_t hash, bucket;
/* Add this entry to the links cache. */
le = calloc(1, sizeof(struct links_entry));
if (le == NULL)
return (NULL);
le->canonical = archive_entry_clone(entry);
/* If the links cache is getting too full, enlarge the hash table. */
if (res->number_entries > res->number_buckets * 2)
grow_hash(res);
hash = (size_t)(archive_entry_dev(entry) ^ archive_entry_ino64(entry));
bucket = hash & (res->number_buckets - 1);
/* If we could allocate the entry, record it. */
if (res->buckets[bucket] != NULL)
res->buckets[bucket]->previous = le;
res->number_entries++;
le->next = res->buckets[bucket];
le->previous = NULL;
res->buckets[bucket] = le;
le->hash = hash;
le->links = archive_entry_nlink(entry) - 1;
return (le);
}
static void
grow_hash(struct archive_entry_linkresolver *res)
{
struct links_entry *le, **new_buckets;
size_t new_size;
size_t i, bucket;
/* Try to enlarge the bucket list. */
new_size = res->number_buckets * 2;
if (new_size < res->number_buckets)
return;
new_buckets = calloc(new_size, sizeof(struct links_entry *));
if (new_buckets == NULL)
return;
for (i = 0; i < res->number_buckets; i++) {
while (res->buckets[i] != NULL) {
/* Remove entry from old bucket. */
le = res->buckets[i];
res->buckets[i] = le->next;
/* Add entry to new bucket. */
bucket = le->hash & (new_size - 1);
if (new_buckets[bucket] != NULL)
new_buckets[bucket]->previous = le;
le->next = new_buckets[bucket];
le->previous = NULL;
new_buckets[bucket] = le;
}
}
free(res->buckets);
res->buckets = new_buckets;
res->number_buckets = new_size;
}
struct archive_entry *
archive_entry_partial_links(struct archive_entry_linkresolver *res,
unsigned int *links)
{
struct archive_entry *e;
struct links_entry *le;
/* Free a held entry. */
if (res->spare != NULL) {
archive_entry_free(res->spare->canonical);
archive_entry_free(res->spare->entry);
free(res->spare);
res->spare = NULL;
}
le = next_entry(res, NEXT_ENTRY_PARTIAL);
if (le != NULL) {
e = le->canonical;
if (links != NULL)
*links = le->links;
le->canonical = NULL;
} else {
e = NULL;
if (links != NULL)
*links = 0;
}
return (e);
}