blob: 433976acd5eefb618deb113993ac15b80d4f22f4 [file] [log] [blame] [raw]
/*
* -----------------------------------------------------------------------------
* "THE BEER-WARE LICENSE" (Revision 42):
* Lukas Niederbremer <webmaster@flippeh.de> and Clark Gaebel <cg.wowus.cg@gmail.com>
* wrote this file. As long as you retain this notice you can do whatever you
* want with this stuff. If we meet some day, and you think this stuff is worth
* it, you can buy us a beer in return.
* -----------------------------------------------------------------------------
*/
#include "nbt.h"
#include <assert.h>
#include <errno.h>
#include <inttypes.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <zlib.h>
/* works around a bug in icc */
int fileno(FILE*);
/* are we running on a little-endian system? */
static inline int little_endian()
{
union {
uint16_t i;
char c[2];
} t = { 0x0001 };
return *t.c == 1;
}
static inline void* swap_bytes(void* s, size_t len)
{
for(char* b = s,
* e = b + len - 1;
b < e;
b++, e--)
{
char t = *b;
*b = *e;
*e = t;
}
return s;
}
/* big endian to native endian. works in-place */
static inline void* be2ne(void* s, size_t len)
{
return little_endian() ? swap_bytes(s, len) : s;
}
/* native endian to big endian. works the exact same as its inverse */
#define ne2be be2ne
/* A special form of memcpy which copies `n' bytes into `dest', then returns
* `src' + n.
*/
static inline const void* memscan(void* dest, const void* src, size_t n)
{
memcpy(dest, src, n);
return (const char*)src + n;
}
/* Does a memscan, then goes from big endian to native endian on the
* destination.
*/
static inline const void* swapped_memscan(void* dest, const void* src, size_t n)
{
const void* ret = memscan(dest, src, n);
return be2ne(dest, n), ret;
}
#define CHECKED_MALLOC(var, n, on_error) do { \
if((var = malloc(n)) == NULL) \
{ \
errno = NBT_EMEM; \
on_error; \
} \
} while(0)
#define CHECKED_GZWRITE(fp, ptr, len) do { \
if(gzwrite((fp), (ptr), (len)) != (int)(len)) \
return NBT_EGZ; \
} while(0)
/* Parses a tag, given a name (may be NULL) and a type. Fills in the payload. */
static nbt_node* parse_unnamed_tag(nbt_type type, char* name, const char** memory, size_t* length);
/*
* Reads some bytes from the memory stream. This macro will read `n'
* bytes into `dest', call either memscan or swapped_memscan depending on
* `scanner', then fix the length. If anything funky goes down, `on_failure'
* will be executed.
*/
#define READ_GENERIC(dest, n, scanner, on_failure) do { \
if(*length < (n)) { on_failure; } \
*memory = scanner((dest), *memory, (n)); \
*length -= (n); \
} while(0)
/*
* Reads a string from memory, moving the pointer and updating the length
* appropriately. Returns NULL on failure.
*/
static inline char* read_string(const char** memory, size_t* length)
{
int16_t string_length;
char* ret = NULL;
READ_GENERIC(&string_length, sizeof string_length, swapped_memscan, goto parse_error);
if(string_length < 0) goto parse_error;
if(*length < (size_t)string_length) goto parse_error;
CHECKED_MALLOC(ret, string_length + 1, goto parse_error);
READ_GENERIC(ret, (size_t)string_length, memscan, goto parse_error);
ret[string_length] = '\0'; /* don't forget to NULL-terminate ;) */
return ret;
parse_error:
if(errno == NBT_OK)
errno = NBT_ERR;
free(ret);
return NULL;
}
static inline struct nbt_byte_array read_byte_array(const char** memory, size_t* length)
{
struct nbt_byte_array ret;
ret.data = NULL;
READ_GENERIC(&ret.length, sizeof ret.length, swapped_memscan, goto parse_error);
if(ret.length < 0) goto parse_error;
CHECKED_MALLOC(ret.data, ret.length, goto parse_error);
READ_GENERIC(ret.data, (size_t)ret.length, memscan, goto parse_error);
return ret;
parse_error:
if(errno == NBT_OK)
errno = NBT_ERR;
free(ret.data);
ret.data = NULL;
return ret;
}
static struct tag_list* read_list(const char** memory, size_t* length)
{
uint8_t type;
int32_t elems;
struct tag_list* ret = NULL;
READ_GENERIC(&type, sizeof type, swapped_memscan, goto parse_error);
READ_GENERIC(&elems, sizeof elems, swapped_memscan, goto parse_error);
CHECKED_MALLOC(ret, sizeof *ret, goto parse_error);
ret->data = NULL; /* the first value in a list is a sentinel. don't even try to read it. */
INIT_LIST_HEAD(&ret->entry);
for(int32_t i = 0; i < elems; i++)
{
struct tag_list* new;
CHECKED_MALLOC(new, sizeof *new, goto parse_error);
new->data = parse_unnamed_tag((nbt_type)type, NULL, memory, length);
if(new->data == NULL)
{
free(new);
goto parse_error;
}
list_add_tail(&new->entry, &ret->entry);
}
return ret;
parse_error:
if(errno == NBT_OK)
errno = NBT_ERR;
nbt_free_list(ret);
return NULL;
}
static struct tag_list* read_compound(const char** memory, size_t* length)
{
struct tag_list* ret;
CHECKED_MALLOC(ret, sizeof *ret, goto parse_error);
ret->data = NULL;
INIT_LIST_HEAD(&ret->entry);
for(;;)
{
uint8_t type;
char* name = NULL;
struct tag_list* new_entry;
READ_GENERIC(&type, sizeof type, swapped_memscan, goto parse_error);
if(type == 0) break; /* TAG_END == 0. We've hit the end of the list when type == TAG_END. */
name = read_string(memory, length);
if(name == NULL) goto parse_error;
CHECKED_MALLOC(new_entry, sizeof *new_entry,
free(name);
goto parse_error;
);
new_entry->data = parse_unnamed_tag((nbt_type)type, name, memory, length);
if(new_entry->data == NULL)
{
free(new_entry);
free(name);
goto parse_error;
}
list_add_tail(&new_entry->entry, &ret->entry);
}
return ret;
parse_error:
if(errno == NBT_OK)
errno = NBT_ERR;
nbt_free_list(ret);
return NULL;
}
/*
* Parses a tag, given a name (may be NULL) and a type. Fills in the payload.
*/
static inline nbt_node* parse_unnamed_tag(nbt_type type, char* name, const char** memory, size_t* length)
{
nbt_node* node;
CHECKED_MALLOC(node, sizeof *node, goto parse_error);
node->type = type;
node->name = name;
#define COPY_INTO_PAYLOAD(payload_name) \
READ_GENERIC(&node->payload.payload_name, sizeof node->payload.payload_name, swapped_memscan, goto parse_error);
switch(type)
{
case TAG_BYTE:
COPY_INTO_PAYLOAD(tag_byte);
break;
case TAG_SHORT:
COPY_INTO_PAYLOAD(tag_short);
break;
case TAG_INT:
COPY_INTO_PAYLOAD(tag_int);
break;
case TAG_LONG:
COPY_INTO_PAYLOAD(tag_long);
break;
case TAG_FLOAT:
COPY_INTO_PAYLOAD(tag_float);
break;
case TAG_DOUBLE:
COPY_INTO_PAYLOAD(tag_double);
break;
case TAG_BYTE_ARRAY:
node->payload.tag_byte_array = read_byte_array(memory, length);
break;
case TAG_STRING:
node->payload.tag_string = read_string(memory, length);
break;
case TAG_LIST:
node->payload.tag_list = read_list(memory, length);
break;
case TAG_COMPOUND:
node->payload.tag_compound = read_compound(memory, length);
break;
default:
goto parse_error; /* Unknown node or TAG_END. Either way, we shouldn't be parsing this. */
}
#undef COPY_INTO_PAYLOAD
if(errno != NBT_OK) goto parse_error;
return node;
parse_error:
if(errno == NBT_OK)
errno = NBT_ERR;
free(node);
return NULL;
}
nbt_node* nbt_parse(const void* mem, size_t len)
{
errno = NBT_OK;
const char** memory = (const char**)&mem;
size_t* length = &len;
/*
* this needs to stay up here since it's referenced by the parse_error
* block.
*/
char* name = NULL;
uint8_t type;
READ_GENERIC(&type, sizeof type, memscan, goto parse_error);
name = read_string(memory, length);
if(name == NULL) goto parse_error;
nbt_node* ret = parse_unnamed_tag((nbt_type)type, name, memory, length);
if(errno != NBT_OK) goto parse_error;
return ret;
parse_error:
if(errno == NBT_OK)
errno = NBT_ERR;
free(name);
return NULL;
}
typedef struct {
char* d; /* data */
size_t l; /* length */
} buffer_t;
/* parses the whole file into a buffer */
static inline buffer_t __parse_file(gzFile fp)
{
buffer_t ret = { NULL, 0 };
/* WARNING: This loop runs in O(n^2). TODO: Fix this! */
for(;;)
{
{ /* resize buffer */
char* temp = realloc(ret.d, ret.l + 4096);
if(temp == NULL) goto parse_error;
ret.d = temp;
}
/* copy in */
size_t bytes_read = gzread(fp, ret.d + ret.l, 4096);
int err;
gzerror(fp, &err);
if(err) { errno = NBT_EGZ; goto parse_error; }
if(bytes_read == 0) break;
/* fix ret.l */
ret.l += bytes_read;
}
return ret;
parse_error:
if(errno == NBT_OK)
errno = NBT_EMEM;
free(ret.d);
ret.d = NULL;
return ret;
}
/*
* No incremental parsing goes on. We just dump the whole decompressed file into
* memory then pass the job off to nbt_parse.
*/
nbt_node* nbt_parse_file(FILE* fp)
{
nbt_node* ret;
errno = NBT_OK;
/*
* We need to keep these declarations up here as opposed to where they're
* used because they're referenced by the parse_error block.
*/
buffer_t buf = { NULL, 0 };
gzFile f = Z_NULL;
if(fp == NULL) return NULL;
int fd = fileno(fp);
if(fd == -1) goto parse_error;
f = gzdopen(fd, "rb");
if(f == Z_NULL) goto parse_error;
buf = __parse_file(f);
if(buf.d == NULL) goto parse_error;
if(gzclose(f) != Z_OK) goto parse_error;
ret = nbt_parse(buf.d, buf.l);
free(buf.d);
return ret;
parse_error:
if(errno == NBT_OK)
errno = NBT_EGZ;
free(buf.d);
if(f != Z_NULL)
gzclose(f);
return NULL;
}
/* spaces, not tabs ;) */
static inline void indent(FILE* fp, size_t amount)
{
for(size_t i = 0; i < amount; i++)
fprintf(fp, " ");
}
static nbt_status __nbt_dump_ascii(const nbt_node* tree, FILE* fp, size_t ident);
/* prints the node's name, or (null) if it has none. */
#define SAFE_NAME(node) ((node)->name ? (node)->name : "<null>")
static inline void dump_byte_array(const struct nbt_byte_array ba, FILE* fp)
{
assert(ba.length >= 0);
fprintf(fp, "[ ");
for(int32_t i = 0; i < ba.length; ++i)
fprintf(fp, "%i ", (int)ba.data[i]);
fprintf(fp, "]");
}
static inline nbt_status dump_list_contents_ascii(const struct tag_list* list, FILE* fp, size_t ident)
{
const struct list_head* pos;
list_for_each(pos, &list->entry)
{
const struct tag_list* entry = list_entry(pos, const struct tag_list, entry);
nbt_status err;
if((err = __nbt_dump_ascii(entry->data, fp, ident)) != NBT_OK)
return err;
}
return NBT_OK;
}
static inline nbt_status __nbt_dump_ascii(const nbt_node* tree, FILE* fp, size_t ident)
{
if(tree == NULL) return NBT_OK;
indent(fp, ident);
if(tree->type == TAG_BYTE)
fprintf(fp, "TAG_Byte(\"%s\"): %i\n", SAFE_NAME(tree), (int)tree->payload.tag_byte);
else if(tree->type == TAG_SHORT)
fprintf(fp, "TAG_Short(\"%s\"): %i\n", SAFE_NAME(tree), (int)tree->payload.tag_short);
else if(tree->type == TAG_INT)
fprintf(fp, "TAG_Int(\"%s\"): %i\n", SAFE_NAME(tree), (int)tree->payload.tag_int);
else if(tree->type == TAG_LONG)
fprintf(fp, "TAG_Long(\"%s\"): %" PRIi64 "\n", SAFE_NAME(tree), tree->payload.tag_long);
else if(tree->type == TAG_FLOAT)
fprintf(fp, "TAG_Float(\"%s\"): %f\n", SAFE_NAME(tree), (double)tree->payload.tag_float);
else if(tree->type == TAG_DOUBLE)
fprintf(fp, "TAG_Double(\"%s\"): %f\n", SAFE_NAME(tree), tree->payload.tag_double);
else if(tree->type == TAG_BYTE_ARRAY)
{
fprintf(fp, "TAG_Byte_Array(\"%s\"): ", SAFE_NAME(tree));
dump_byte_array(tree->payload.tag_byte_array, fp);
fprintf(fp, "\n");
}
else if(tree->type == TAG_STRING)
{
if(tree->payload.tag_string == NULL)
return NBT_ERR;
fprintf(fp, "TAG_String(\"%s\"): %s\n", SAFE_NAME(tree), tree->payload.tag_string);
}
else if(tree->type == TAG_LIST)
{
fprintf(fp, "TAG_List(\"%s\")\n", SAFE_NAME(tree));
indent(fp, ident);
fprintf(fp, "{\n");
nbt_status err;
if((err = dump_list_contents_ascii(tree->payload.tag_list, fp, ident + 1)) != NBT_OK)
return err;
indent(fp, ident);
fprintf(fp, "}\n");
}
else if(tree->type == TAG_COMPOUND)
{
fprintf(fp, "TAG_Compound(\"%s\")\n", SAFE_NAME(tree));
indent(fp, ident);
fprintf(fp, "{\n");
nbt_status err;
if((err = dump_list_contents_ascii(tree->payload.tag_compound, fp, ident + 1)) != NBT_OK)
return err;
indent(fp, ident);
fprintf(fp, "}\n");
}
else
return NBT_ERR;
return NBT_OK;
}
nbt_status nbt_dump_ascii(const nbt_node* tree, FILE* fp)
{
return __nbt_dump_ascii(tree, fp, 0);
}
static nbt_status dump_byte_array_binary(const struct nbt_byte_array ba, gzFile fp)
{
int32_t dumped_length = ba.length;
ne2be(&dumped_length, sizeof dumped_length);
CHECKED_GZWRITE(fp, &dumped_length, sizeof dumped_length);
if(ba.length) assert(ba.data);
CHECKED_GZWRITE(fp, ba.data, ba.length);
return NBT_OK;
}
static nbt_status dump_string_binary(const char* name, gzFile fp)
{
assert(name);
size_t len = strlen(name);
if(len > 32767 /* SHORT_MAX */)
return NBT_ERR;
{ /* dump the length */
int16_t dumped_len = (int16_t)len;
ne2be(&dumped_len, sizeof dumped_len);
CHECKED_GZWRITE(fp, &dumped_len, sizeof dumped_len);
}
CHECKED_GZWRITE(fp, name, len);
return NBT_OK;
}
/*
* Is the list all one type? If yes, return the type. Otherwise, return
* TAG_INVALID
*/
static inline nbt_type list_is_homogenous(const struct tag_list* list)
{
nbt_type type = TAG_INVALID;
const struct list_head* pos;
list_for_each(pos, &list->entry)
{
const struct tag_list* cur = list_entry(pos, const struct tag_list, entry);
assert(cur->data->type != TAG_INVALID);
if(cur->data->type == TAG_INVALID)
return TAG_INVALID;
/* if we're the first type, just set it to our current type */
if(type == TAG_INVALID) type = cur->data->type;
if(type != cur->data->type)
return TAG_INVALID;
}
return type;
}
static nbt_status __dump_binary(const nbt_node*, bool, gzFile);
static nbt_status dump_list_binary(const struct tag_list* list, gzFile fp)
{
nbt_type type;
size_t len = list_length(&list->entry);
if(len == 0) /* empty lists can just be silently ignored */
return NBT_OK;
if(len > 2147483647 /* INT_MAX */)
return NBT_ERR;
assert(list_is_homogenous(list) != TAG_INVALID);
if((type = list_is_homogenous(list)) == TAG_INVALID)
return NBT_ERR;
{
int8_t _type = (int8_t)type;
ne2be(&_type, sizeof _type); /* unnecessary, but left in to keep similar code looking similar */
CHECKED_GZWRITE(fp, &_type, sizeof _type);
}
{
int32_t dumped_len = (int32_t)len;
ne2be(&dumped_len, sizeof dumped_len);
CHECKED_GZWRITE(fp, &dumped_len, sizeof dumped_len);
}
const struct list_head* pos;
list_for_each(pos, &list->entry)
{
const struct tag_list* entry = list_entry(pos, const struct tag_list, entry);
nbt_status ret;
if((ret = __dump_binary(entry->data, false, fp)) != NBT_OK)
return ret;
}
return NBT_OK;
}
static nbt_status dump_compound_binary(const struct tag_list* list, gzFile fp)
{
if(list_empty(&list->entry)) /* empty lists can just be silently ignored */
return NBT_OK;
const struct list_head* pos;
list_for_each(pos, &list->entry)
{
const struct tag_list* entry = list_entry(pos, const struct tag_list, entry);
nbt_status ret;
if((ret = __dump_binary(entry->data, true, fp)) != NBT_OK)
return ret;
}
/* write out TAG_End */
uint8_t zero = 0;
CHECKED_GZWRITE(fp, &zero, sizeof zero);
return NBT_OK;
}
/*
* @param dump_type Should we dump the type, or just skip it? We need to skip
* when dumping lists, because the list header already says
* the type.
*/
static inline nbt_status __dump_binary(const nbt_node* tree, bool dump_type, gzFile fp)
{
if(dump_type)
{ /* write out the type */
int8_t type = (int8_t)tree->type;
CHECKED_GZWRITE(fp, &type, sizeof type);
}
if(tree->name)
{
nbt_status err;
if((err = dump_string_binary(tree->name, fp)) != NBT_OK)
return err;
}
#define DUMP_NUM(type, x) do { \
type temp = x; \
ne2be(&temp, sizeof temp); \
CHECKED_GZWRITE(fp, &temp, sizeof temp); \
} while(0)
if(tree->type == TAG_BYTE)
DUMP_NUM(int8_t, tree->payload.tag_byte);
else if(tree->type == TAG_SHORT)
DUMP_NUM(int16_t, tree->payload.tag_short);
else if(tree->type == TAG_INT)
DUMP_NUM(int32_t, tree->payload.tag_int);
else if(tree->type == TAG_LONG)
DUMP_NUM(int64_t, tree->payload.tag_long);
else if(tree->type == TAG_FLOAT)
DUMP_NUM(float, tree->payload.tag_float);
else if(tree->type == TAG_DOUBLE)
DUMP_NUM(double, tree->payload.tag_double);
else if(tree->type == TAG_BYTE_ARRAY)
return dump_byte_array_binary(tree->payload.tag_byte_array, fp);
else if(tree->type == TAG_STRING)
return dump_string_binary(tree->payload.tag_string, fp);
else if(tree->type == TAG_LIST)
return dump_list_binary(tree->payload.tag_list, fp);
else if(tree->type == TAG_COMPOUND)
return dump_compound_binary(tree->payload.tag_compound, fp);
else
return NBT_ERR;
return NBT_OK;
#undef DUMP_NUM
}
nbt_status nbt_dump_binary(const nbt_node* tree, FILE* fp)
{
if(tree == NULL) return NBT_OK;
int fd = fileno(fp);
if(fd == -1) return NBT_EGZ;
gzFile f = gzdopen(fd, "wb");
if(f == Z_NULL) return NBT_EGZ;
nbt_status r = __dump_binary(tree, true, f);
gzclose(f);
return r;
}