mirror of https://github.com/sipwise/rtpengine.git
You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
605 lines
26 KiB
605 lines
26 KiB
#ifndef _BENCODE_H_
|
|
#define _BENCODE_H_
|
|
|
|
#include <sys/uio.h>
|
|
#include <string.h>
|
|
#include <json-glib/json-glib.h>
|
|
|
|
#include "compat.h"
|
|
|
|
struct bencode_buffer;
|
|
enum bencode_type;
|
|
struct bencode_item;
|
|
struct __bencode_buffer_piece;
|
|
struct __bencode_free_list;
|
|
|
|
typedef enum bencode_type bencode_type_t;
|
|
typedef struct bencode_buffer bencode_buffer_t;
|
|
typedef struct bencode_item bencode_item_t;
|
|
typedef void (*free_func_t)(void *);
|
|
|
|
enum bencode_type {
|
|
BENCODE_INVALID = 0,
|
|
BENCODE_STRING, /* byte string */
|
|
BENCODE_INTEGER, /* long long int */
|
|
BENCODE_LIST, /* flat list of other objects */
|
|
BENCODE_DICTIONARY, /* dictionary of key/values pairs. keys are always strings */
|
|
BENCODE_END_MARKER, /* used internally only */
|
|
};
|
|
|
|
struct bencode_item {
|
|
bencode_type_t type;
|
|
struct iovec iov[2]; /* when decoding, iov[1] contains the contents of a string object */
|
|
unsigned int iov_cnt;
|
|
size_t str_len; /* length of the whole ENCODED object. NOT the length of a byte string */
|
|
long long int value; /* when decoding an integer, contains the value; otherwise used internally */
|
|
bencode_item_t *parent, *child, *last_child, *sibling;
|
|
bencode_buffer_t *buffer;
|
|
char __buf[0];
|
|
};
|
|
|
|
struct bencode_buffer {
|
|
struct __bencode_buffer_piece *pieces;
|
|
struct __bencode_free_list *free_list;
|
|
unsigned int error:1; /* set to !0 if allocation failed at any point */
|
|
};
|
|
|
|
|
|
|
|
|
|
|
|
/* to embed BENCODE_STRING objects into printf-like functions */
|
|
#define BENCODE_FORMAT "%.*s"
|
|
#define BENCODE_FMT(b) (int) (b)->iov[1].iov_len, (char *) (b)->iov[1].iov_base
|
|
|
|
|
|
|
|
|
|
/*** INIT & DESTROY ***/
|
|
|
|
/* Initializes a bencode_buffer_t object. This object is used to group together all memory allocations
|
|
* made when encoding or decoding. Its memory usage is always growing, until it is freed, at which point
|
|
* all objects created through it become invalid. The actual object must be allocated separately, for
|
|
* example by being put on the stack.
|
|
* Returns 0 on success or -1 on failure (if no memory could be allocated). */
|
|
int bencode_buffer_init(bencode_buffer_t *buf);
|
|
|
|
/* Allocate a piece of memory from the given buffer object */
|
|
void *bencode_buffer_alloc(bencode_buffer_t *, unsigned int);
|
|
|
|
/* Destroys a previously initialized bencode_buffer_t object. All memory used by the object is freed
|
|
* and all objects created through it become invalid. */
|
|
void bencode_buffer_free(bencode_buffer_t *buf);
|
|
|
|
/* Creates a new empty dictionary object. Memory will be allocated from the bencode_buffer_t object.
|
|
* Returns NULL if no memory could be allocated. */
|
|
bencode_item_t *bencode_dictionary(bencode_buffer_t *buf);
|
|
|
|
/* Creates a new empty list object. Memory will be allocated from the bencode_buffer_t object.
|
|
* Returns NULL if no memory could be allocated. */
|
|
bencode_item_t *bencode_list(bencode_buffer_t *buf);
|
|
|
|
/* Adds a pointer to the bencode_buffer_t object's internal free list. When the bencode_buffer_t
|
|
* object is destroyed, the specified function will be called on this pointer. */
|
|
void bencode_buffer_destroy_add(bencode_buffer_t *buf, free_func_t, void *);
|
|
|
|
/* Returns the buffer associated with an item, or NULL if pointer given is NULL */
|
|
INLINE bencode_buffer_t *bencode_item_buffer(bencode_item_t *);
|
|
|
|
/* like strdup() but uses the bencode buffer to store the string */
|
|
INLINE char *bencode_strdup(bencode_buffer_t *, const char *);
|
|
|
|
/* ditto but inits a str object */
|
|
INLINE void bencode_strdup_str(bencode_buffer_t *, str *, const char *);
|
|
|
|
|
|
|
|
|
|
|
|
/*** DICTIONARY BUILDING ***/
|
|
|
|
/* Adds a new key/value pair to a dictionary. Memory will be allocated from the same bencode_buffer_t
|
|
* object as the dictionary was allocated from. Returns NULL if no memory could be allocated, otherwise
|
|
* returns "val".
|
|
* The function does not check whether the key being added is already present in the dictionary.
|
|
* Also, the function does not reorder keys into lexicographical order; keys will be encoded in
|
|
* the same order as they've been added. The key must a null-terminated string.
|
|
* The value to be added must not have been previously linked into any other dictionary or list. */
|
|
INLINE bencode_item_t *bencode_dictionary_add(bencode_item_t *dict, const char *key, bencode_item_t *val);
|
|
INLINE bencode_item_t *bencode_dictionary_str_add(bencode_item_t *dict, const str *key, bencode_item_t *val);
|
|
|
|
/* Identical to bencode_dictionary_add() but doesn't require the key string to be null-terminated */
|
|
bencode_item_t *bencode_dictionary_add_len(bencode_item_t *dict, const char *key, size_t keylen, bencode_item_t *val);
|
|
|
|
/* Convenience function to add a string value to a dictionary, possibly duplicated into the
|
|
* bencode_buffer_t object. */
|
|
INLINE bencode_item_t *bencode_dictionary_add_string(bencode_item_t *dict, const char *key, const char *val);
|
|
INLINE bencode_item_t *bencode_dictionary_add_string_dup(bencode_item_t *dict, const char *key, const char *val);
|
|
INLINE bencode_item_t *bencode_dictionary_add_string_len(bencode_item_t *dict, const char *key, const char *val, size_t len);
|
|
|
|
/* Ditto, but for a "str" object */
|
|
INLINE bencode_item_t *bencode_dictionary_add_str(bencode_item_t *dict, const char *key, const str *val);
|
|
INLINE bencode_item_t *bencode_dictionary_str_add_str(bencode_item_t *dict, const str *key, const str *val);
|
|
INLINE bencode_item_t *bencode_dictionary_add_str_dup(bencode_item_t *dict, const char *key, const str *val);
|
|
|
|
/* Convenience functions to add the respective (newly created) objects to a dictionary */
|
|
INLINE bencode_item_t *bencode_dictionary_add_integer(bencode_item_t *dict, const char *key, long long int val);
|
|
INLINE bencode_item_t *bencode_dictionary_add_dictionary(bencode_item_t *dict, const char *key);
|
|
INLINE bencode_item_t *bencode_dictionary_add_list(bencode_item_t *dict, const char *key);
|
|
|
|
|
|
|
|
|
|
|
|
/*** LIST BUILDING ***/
|
|
|
|
/* Adds a new item to a list. Returns "item".
|
|
* The item to be added must not have been previously linked into any other dictionary or list. */
|
|
bencode_item_t *bencode_list_add(bencode_item_t *list, bencode_item_t *item);
|
|
|
|
/* Convenience function to add the respective (newly created) objects to a list */
|
|
INLINE bencode_item_t *bencode_list_add_string(bencode_item_t *list, const char *s);
|
|
INLINE bencode_item_t *bencode_list_add_string_dup(bencode_item_t *list, const char *s);
|
|
INLINE bencode_item_t *bencode_list_add_str(bencode_item_t *list, const str *s);
|
|
INLINE bencode_item_t *bencode_list_add_str_dup(bencode_item_t *list, const str *s);
|
|
INLINE bencode_item_t *bencode_list_add_list(bencode_item_t *list);
|
|
INLINE bencode_item_t *bencode_list_add_dictionary(bencode_item_t *list);
|
|
|
|
|
|
|
|
|
|
|
|
/*** STRING BUILDING & HANDLING ***/
|
|
|
|
/* Creates a new byte-string object. The given string does not have to be null-terminated, instead
|
|
* the length of the string is specified by the "len" parameter. Returns NULL if no memory could
|
|
* be allocated.
|
|
* Strings are not copied or duplicated, so the string pointed to by "s" must remain valid until
|
|
* the complete document is finally encoded or sent out. */
|
|
bencode_item_t *bencode_string_len(bencode_buffer_t *buf, const char *s, size_t len);
|
|
|
|
/* Creates a new byte-string object. The given string must be null-terminated. Otherwise identical
|
|
* to bencode_string_len(). */
|
|
INLINE bencode_item_t *bencode_string(bencode_buffer_t *buf, const char *s);
|
|
|
|
/* Creates a new byte-string object from a "str" object. The string does not have to be null-
|
|
* terminated. */
|
|
INLINE bencode_item_t *bencode_str(bencode_buffer_t *buf, const str *s);
|
|
|
|
/* Identical to the above three functions, but copies the string into the bencode_buffer_t object.
|
|
* Thus, the given string doesn't have to remain valid and accessible afterwards. */
|
|
bencode_item_t *bencode_string_len_dup(bencode_buffer_t *buf, const char *s, size_t len);
|
|
INLINE bencode_item_t *bencode_string_dup(bencode_buffer_t *buf, const char *s);
|
|
INLINE bencode_item_t *bencode_str_dup(bencode_buffer_t *buf, const str *s);
|
|
|
|
/* Convenience function to compare a string object to a regular C string. Returns 2 if object
|
|
* isn't a string object, otherwise returns according to strcmp(). */
|
|
INLINE int bencode_strcmp(bencode_item_t *a, const char *b);
|
|
|
|
/* Converts the string object "in" into a str object "out". Returns "out" on success, or NULL on
|
|
* error ("in" was NULL or not a string object). */
|
|
INLINE str *bencode_get_str(bencode_item_t *in, str *out);
|
|
|
|
|
|
|
|
|
|
|
|
/*** INTEGER BUILDING ***/
|
|
|
|
/* Creates a new integer object. Returns NULL if no memory could be allocated. */
|
|
bencode_item_t *bencode_integer(bencode_buffer_t *buf, long long int i);
|
|
|
|
// Return integer, possibly converted from string
|
|
INLINE long long bencode_get_integer_str(bencode_item_t *item, long long int defval);
|
|
|
|
|
|
|
|
|
|
|
|
/*** COLLAPSING & ENCODING ***/
|
|
|
|
/* Collapses and encodes the complete document structure under the "root" element (which normally
|
|
* is either a dictionary or a list) into an array of "iovec" structures. This array can then be
|
|
* passed to functions ala writev() or sendmsg() to output the encoded document as a whole. Memory
|
|
* is allocated from the same bencode_buffer_t object as the "root" object was allocated from.
|
|
* The "head" and "tail" parameters specify additional "iovec" structures that should be included
|
|
* in the allocated array before or after (respectively) the iovec structures used by the encoded
|
|
* document. Both parameters can be zero if no additional elements in the array are required.
|
|
* Returns a pointer to the allocated array or NULL if no memory could be allocated. The number of
|
|
* array elements is returned in "cnt" which must be a valid pointer to an int. This number does
|
|
* not include any additional elements allocated through the "head" or "tail" parameters.
|
|
*
|
|
* Therefore, the contents of the returned array are:
|
|
* [0 .. (head - 1)] = unused and uninitialized iovec structures
|
|
* [(head) .. (head + cnt - 1)] = the encoded document
|
|
* [(head + cnt) .. (head + cnt + tail - 1)] = unused and uninitialized iovec structures
|
|
*
|
|
* The returned array will be freed when the corresponding bencode_buffer_t object is destroyed. */
|
|
struct iovec *bencode_iovec(bencode_item_t *root, int *cnt, unsigned int head, unsigned int tail);
|
|
|
|
/* Similar to bencode_iovec(), but instead returns the encoded document as a null-terminated string.
|
|
* Memory for the string is allocated from the same bencode_buffer_t object as the "root" object
|
|
* was allocated from. If "len" is a non-NULL pointer, the length of the generated string is returned
|
|
* in *len. This is important if the encoded document contains binary data, in which case null
|
|
* termination cannot be trusted. The returned string is freed when the corresponding
|
|
* bencode_buffer_t object is destroyed. */
|
|
char *bencode_collapse(bencode_item_t *root, size_t *len);
|
|
|
|
/* Identical to bencode_collapse() but fills in a "str" object. Returns "out". */
|
|
INLINE str *bencode_collapse_str(bencode_item_t *root, str *out);
|
|
|
|
/* Identical to bencode_collapse(), but the memory for the returned string is not allocated from
|
|
* a bencode_buffer_t object, but instead using the function defined as BENCODE_MALLOC (normally
|
|
* malloc() or pkg_malloc()), similar to strdup(). Using this function, the bencode_buffer_t
|
|
* object can be destroyed, but the returned string remains valid and usable. */
|
|
char *bencode_collapse_dup(bencode_item_t *root, size_t *len);
|
|
|
|
// Collapse into a JSON document. Otherwise identical to bencode_collapse_str.
|
|
str *bencode_collapse_str_json(bencode_item_t *root, str *out);
|
|
|
|
|
|
|
|
|
|
|
|
/*** DECODING ***/
|
|
|
|
/* Decodes an encoded document from a string into a tree of bencode_item_t objects. The string does
|
|
* not need to be null-terminated, instead the length of the string is given through the "len"
|
|
* parameter. Memory is allocated from the bencode_buffer_t object. Returns NULL if no memory could
|
|
* be allocated or if the document could not be successfully decoded.
|
|
*
|
|
* The returned element is the "root" of the document tree and normally is either a list object or
|
|
* a dictionary object, but can also be a single string or integer object with no other objects
|
|
* underneath or besides it (no childred and no siblings). The type of the object can be determined
|
|
* by its ->type property.
|
|
*
|
|
* The number of bytes that could successfully be decoded into an object tree can be accessed through
|
|
* the root element's ->str_len property. Normally, this number should be equal to the "len" parameter
|
|
* passed, in which case the full string could be decoded. If ->str_len is less than "len", then there
|
|
* was additional stray byte data after the end of the encoded document.
|
|
*
|
|
* The document tree can be traversed through the ->child and ->sibling pointers in each object. The
|
|
* ->child pointer will be NULL for string and integer objects, as they don't contain other objects.
|
|
* For lists and dictionaries, ->child will be a pointer to the first contained object. This first
|
|
* contained object's ->sibling pointer will point to the next (second) contained object of the list
|
|
* or the dictionary, and so on. The last contained element of a list of dictionary will have a
|
|
* NULL ->sibling pointer.
|
|
*
|
|
* Dictionaries are like lists with ordered key/value pairs. When traversing dictionaries like
|
|
* lists, the following applies: The first element in the list (where ->child points to) will be the
|
|
* key of the first key/value pair (guaranteed to be a string and guaranteed to be present). The
|
|
* next element (following one ->sibling) will be the value of the first key/value pair. Following
|
|
* another ->sibling will point to the key of the next (second) key/value pair, and so on.
|
|
*
|
|
* However, to access children objects of dictionaries, the special functions following the naming
|
|
* scheme bencode_dictionary_get_* below should be used. They perform key lookup through a simple
|
|
* hash built into the dictionary object and so perform the lookup much faster. Only dictionaries
|
|
* created through a decoding process (i.e. not ones created from bencode_dictionary()) have this
|
|
* property. The hash is efficient only up to a certain number of elements (BENCODE_HASH_BUCKETS
|
|
* in bencode.c) contained in the dictionary. If the number of children object exceeds this number,
|
|
* key lookup will be slower than simply linearly traversing the list.
|
|
*
|
|
* The decoding function for dictionary object does not check whether keys are unique within the
|
|
* dictionary. It also does not care about lexicographical order of the keys.
|
|
*
|
|
* Decoded string objects will contain the raw decoded byte string in ->iov[1] (including the correct
|
|
* length). Strings are NOT null-terminated. Decoded integer objects will contain the decoded value
|
|
* in ->value.
|
|
*
|
|
* All memory is freed when the bencode_buffer_t object is destroyed.
|
|
*/
|
|
bencode_item_t *bencode_decode(bencode_buffer_t *buf, const char *s, size_t len);
|
|
|
|
/* Identical to bencode_decode(), but returns successfully only if the type of the decoded object match
|
|
* "expect". */
|
|
INLINE bencode_item_t *bencode_decode_expect(bencode_buffer_t *buf, const char *s, size_t len, bencode_type_t expect);
|
|
|
|
/* Identical to bencode_decode_expect() but takes a "str" argument. */
|
|
INLINE bencode_item_t *bencode_decode_expect_str(bencode_buffer_t *buf, const str *s, bencode_type_t expect);
|
|
|
|
/* Returns the number of bytes that could successfully be decoded from 's', -1 if more bytes are needed or -2 on error */
|
|
ssize_t bencode_valid(const char *s, size_t len);
|
|
|
|
// Convert a GLib JSON document to bencode
|
|
bencode_item_t *bencode_convert_json(bencode_buffer_t *buf, JsonParser *json);
|
|
|
|
|
|
/*** DICTIONARY LOOKUP & EXTRACTION ***/
|
|
|
|
/* Searches the given dictionary object for the given key and returns the respective value. Returns
|
|
* NULL if the given object isn't a dictionary or if the key doesn't exist. The key must be a
|
|
* null-terminated string. */
|
|
INLINE bencode_item_t *bencode_dictionary_get(bencode_item_t *dict, const char *key);
|
|
|
|
/* Identical to bencode_dictionary_get() but doesn't require the key to be null-terminated. */
|
|
bencode_item_t *bencode_dictionary_get_len(bencode_item_t *dict, const char *key, size_t key_len);
|
|
|
|
/* Identical to bencode_dictionary_get() but returns the value only if its type is a string, and
|
|
* returns it as a pointer to the string itself. Returns NULL if the value is of some other type. The
|
|
* returned string is NOT null-terminated. Length of the string is returned in *len, which must be a
|
|
* valid pointer. The returned string will be valid until dict's bencode_buffer_t object is destroyed. */
|
|
INLINE char *bencode_dictionary_get_string(bencode_item_t *dict, const char *key, size_t *len);
|
|
|
|
/* Identical to bencode_dictionary_get_string() but fills in a "str" struct. Returns str->s, which
|
|
* may be NULL. */
|
|
INLINE char *bencode_dictionary_get_str(bencode_item_t *dict, const char *key, str *str);
|
|
|
|
/* Looks up the given key in the dictionary and compares the corresponding value to the given
|
|
* null-terminated string. Returns 2 if the key isn't found or if the value isn't a string, otherwise
|
|
* returns according to strcmp(). */
|
|
INLINE int bencode_dictionary_get_strcmp(bencode_item_t *dict, const char *key, const char *str);
|
|
|
|
/* Identical to bencode_dictionary_get() but returns the string in a newly allocated buffer (using the
|
|
* BENCODE_MALLOC function), which remains valid even after bencode_buffer_t is destroyed. */
|
|
INLINE char *bencode_dictionary_get_string_dup(bencode_item_t *dict, const char *key, size_t *len);
|
|
|
|
/* Combines bencode_dictionary_get_str() and bencode_dictionary_get_string_dup(). Fills in a "str"
|
|
* struct, but copies the string into a newly allocated buffer. Returns str->s. */
|
|
INLINE char *bencode_dictionary_get_str_dup(bencode_item_t *dict, const char *key, str *str);
|
|
|
|
/* Identical to bencode_dictionary_get_string() but expects an integer object. The parameter "defval"
|
|
* specified which value should be returned if the key is not found or if the value is not an integer. */
|
|
INLINE long long int bencode_dictionary_get_integer(bencode_item_t *dict, const char *key, long long int defval);
|
|
|
|
/* Identical to bencode_dictionary_get_integer() but allows for the item to be a string. */
|
|
INLINE long long int bencode_dictionary_get_int_str(bencode_item_t *dict, const char *key, long long int defval);
|
|
|
|
/* Identical to bencode_dictionary_get(), but returns the object only if its type matches "expect". */
|
|
INLINE bencode_item_t *bencode_dictionary_get_expect(bencode_item_t *dict, const char *key, bencode_type_t expect);
|
|
|
|
|
|
|
|
|
|
|
|
/**************************/
|
|
|
|
INLINE bencode_buffer_t *bencode_item_buffer(bencode_item_t *i) {
|
|
if (!i)
|
|
return NULL;
|
|
return i->buffer;
|
|
}
|
|
|
|
INLINE bencode_item_t *bencode_string(bencode_buffer_t *buf, const char *s) {
|
|
return bencode_string_len(buf, s, strlen(s));
|
|
}
|
|
|
|
INLINE bencode_item_t *bencode_string_dup(bencode_buffer_t *buf, const char *s) {
|
|
return bencode_string_len_dup(buf, s, strlen(s));
|
|
}
|
|
|
|
INLINE bencode_item_t *bencode_str(bencode_buffer_t *buf, const str *s) {
|
|
return bencode_string_len(buf, s->s, s->len);
|
|
}
|
|
|
|
INLINE bencode_item_t *bencode_str_dup(bencode_buffer_t *buf, const str *s) {
|
|
return bencode_string_len_dup(buf, s->s, s->len);
|
|
}
|
|
|
|
INLINE char *bencode_strdup(bencode_buffer_t *buf, const char *s) {
|
|
char *ret = bencode_buffer_alloc(buf, strlen(s) + 1);
|
|
strcpy(ret, s);
|
|
return ret;
|
|
}
|
|
|
|
INLINE void bencode_strdup_str(bencode_buffer_t *buf, str *o, const char *s) {
|
|
o->len = strlen(s);
|
|
o->s = bencode_buffer_alloc(buf, o->len);
|
|
memcpy(o->s, s, o->len);
|
|
}
|
|
|
|
INLINE bencode_item_t *bencode_dictionary_add(bencode_item_t *dict, const char *key, bencode_item_t *val) {
|
|
if (!key)
|
|
return NULL;
|
|
return bencode_dictionary_add_len(dict, key, strlen(key), val);
|
|
}
|
|
|
|
INLINE bencode_item_t *bencode_dictionary_str_add(bencode_item_t *dict, const str *key, bencode_item_t *val) {
|
|
if (!key)
|
|
return NULL;
|
|
return bencode_dictionary_add_len(dict, key->s, key->len, val);
|
|
}
|
|
|
|
INLINE bencode_item_t *bencode_dictionary_add_string(bencode_item_t *dict, const char *key, const char *val) {
|
|
if (!val)
|
|
return NULL;
|
|
return bencode_dictionary_add(dict, key, bencode_string(bencode_item_buffer(dict), val));
|
|
}
|
|
|
|
INLINE bencode_item_t *bencode_dictionary_add_string_len(bencode_item_t *dict, const char *key, const char *val, size_t len) {
|
|
if (!val)
|
|
return NULL;
|
|
return bencode_dictionary_add(dict, key, bencode_string_len(bencode_item_buffer(dict), val, len));
|
|
}
|
|
|
|
INLINE bencode_item_t *bencode_dictionary_add_string_dup(bencode_item_t *dict, const char *key, const char *val) {
|
|
if (!val)
|
|
return NULL;
|
|
return bencode_dictionary_add(dict, key, bencode_string_dup(bencode_item_buffer(dict), val));
|
|
}
|
|
|
|
INLINE bencode_item_t *bencode_dictionary_add_str(bencode_item_t *dict, const char *key, const str *val) {
|
|
if (!val)
|
|
return NULL;
|
|
return bencode_dictionary_add(dict, key, bencode_str(bencode_item_buffer(dict), val));
|
|
}
|
|
|
|
INLINE bencode_item_t *bencode_dictionary_str_add_str(bencode_item_t *dict, const str *key, const str *val) {
|
|
if (!val)
|
|
return NULL;
|
|
return bencode_dictionary_str_add(dict, key, bencode_str(bencode_item_buffer(dict), val));
|
|
}
|
|
|
|
INLINE bencode_item_t *bencode_dictionary_add_str_dup(bencode_item_t *dict, const char *key, const str *val) {
|
|
if (!val)
|
|
return NULL;
|
|
return bencode_dictionary_add(dict, key, bencode_str_dup(bencode_item_buffer(dict), val));
|
|
}
|
|
|
|
INLINE bencode_item_t *bencode_dictionary_add_integer(bencode_item_t *dict, const char *key, long long int val) {
|
|
return bencode_dictionary_add(dict, key, bencode_integer(bencode_item_buffer(dict), val));
|
|
}
|
|
|
|
INLINE bencode_item_t *bencode_dictionary_add_dictionary(bencode_item_t *dict, const char *key) {
|
|
return bencode_dictionary_add(dict, key, bencode_dictionary(bencode_item_buffer(dict)));
|
|
}
|
|
|
|
INLINE bencode_item_t *bencode_dictionary_add_list(bencode_item_t *dict, const char *key) {
|
|
return bencode_dictionary_add(dict, key, bencode_list(bencode_item_buffer(dict)));
|
|
}
|
|
|
|
INLINE bencode_item_t *bencode_list_add_string(bencode_item_t *list, const char *s) {
|
|
return bencode_list_add(list, bencode_string(bencode_item_buffer(list), s));
|
|
}
|
|
|
|
INLINE bencode_item_t *bencode_list_add_string_dup(bencode_item_t *list, const char *s) {
|
|
if (!s)
|
|
return NULL;
|
|
return bencode_list_add(list, bencode_string_dup(bencode_item_buffer(list), s));
|
|
}
|
|
|
|
INLINE bencode_item_t *bencode_list_add_str(bencode_item_t *list, const str *s) {
|
|
return bencode_list_add(list, bencode_str(bencode_item_buffer(list), s));
|
|
}
|
|
|
|
INLINE bencode_item_t *bencode_list_add_str_dup(bencode_item_t *list, const str *s) {
|
|
if (!s)
|
|
return NULL;
|
|
return bencode_list_add(list, bencode_str_dup(bencode_item_buffer(list), s));
|
|
}
|
|
|
|
INLINE bencode_item_t *bencode_list_add_list(bencode_item_t *list) {
|
|
return bencode_list_add(list, bencode_list(bencode_item_buffer(list)));
|
|
}
|
|
|
|
INLINE bencode_item_t *bencode_list_add_dictionary(bencode_item_t *list) {
|
|
return bencode_list_add(list, bencode_dictionary(bencode_item_buffer(list)));
|
|
}
|
|
|
|
INLINE bencode_item_t *bencode_dictionary_get(bencode_item_t *dict, const char *key) {
|
|
if (!key)
|
|
return NULL;
|
|
return bencode_dictionary_get_len(dict, key, strlen(key));
|
|
}
|
|
|
|
INLINE char *bencode_dictionary_get_string(bencode_item_t *dict, const char *key, size_t *len) {
|
|
bencode_item_t *val;
|
|
val = bencode_dictionary_get(dict, key);
|
|
if (!val || val->type != BENCODE_STRING)
|
|
return NULL;
|
|
*len = val->iov[1].iov_len;
|
|
return val->iov[1].iov_base;
|
|
}
|
|
|
|
INLINE char *bencode_dictionary_get_str(bencode_item_t *dict, const char *key, str *s) {
|
|
s->s = bencode_dictionary_get_string(dict, key, &s->len);
|
|
if (!s->s)
|
|
s->len = 0;
|
|
return s->s;
|
|
}
|
|
|
|
INLINE char *bencode_dictionary_get_string_dup(bencode_item_t *dict, const char *key, size_t *len) {
|
|
const char *s;
|
|
char *ret;
|
|
s = bencode_dictionary_get_string(dict, key, len);
|
|
if (!s)
|
|
return NULL;
|
|
ret = BENCODE_MALLOC(*len);
|
|
if (!ret)
|
|
return NULL;
|
|
memcpy(ret, s, *len);
|
|
return ret;
|
|
}
|
|
|
|
INLINE char *bencode_dictionary_get_str_dup(bencode_item_t *dict, const char *key, str *s) {
|
|
s->s = bencode_dictionary_get_string_dup(dict, key, &s->len);
|
|
return s->s;
|
|
}
|
|
|
|
INLINE long long int bencode_dictionary_get_integer(bencode_item_t *dict, const char *key, long long int defval) {
|
|
bencode_item_t *val;
|
|
val = bencode_dictionary_get(dict, key);
|
|
if (!val || val->type != BENCODE_INTEGER)
|
|
return defval;
|
|
return val->value;
|
|
}
|
|
|
|
INLINE long long bencode_get_integer_str(bencode_item_t *val, long long int defval) {
|
|
if (!val)
|
|
return defval;
|
|
if (val->type == BENCODE_INTEGER)
|
|
return val->value;
|
|
if (val->type != BENCODE_STRING)
|
|
return defval;
|
|
if (val->iov[1].iov_len == 0)
|
|
return defval;
|
|
|
|
// uh oh...
|
|
char *s = val->iov[1].iov_base;
|
|
char old = s[val->iov[1].iov_len];
|
|
s[val->iov[1].iov_len] = 0;
|
|
char *errp;
|
|
long long int ret = strtoll(val->iov[1].iov_base, &errp, 10);
|
|
s[val->iov[1].iov_len] = old;
|
|
if (errp != val->iov[1].iov_base + val->iov[1].iov_len)
|
|
return defval;
|
|
return ret;
|
|
}
|
|
|
|
INLINE long long int bencode_dictionary_get_int_str(bencode_item_t *dict, const char *key, long long int defval) {
|
|
bencode_item_t *val;
|
|
val = bencode_dictionary_get(dict, key);
|
|
return bencode_get_integer_str(val, defval);
|
|
}
|
|
|
|
INLINE bencode_item_t *bencode_decode_expect(bencode_buffer_t *buf, const char *s, size_t len, bencode_type_t expect) {
|
|
bencode_item_t *ret;
|
|
ret = bencode_decode(buf, s, len);
|
|
if (!ret || ret->type != expect)
|
|
return NULL;
|
|
return ret;
|
|
}
|
|
|
|
INLINE bencode_item_t *bencode_decode_expect_str(bencode_buffer_t *buf, const str *s, bencode_type_t expect) {
|
|
return bencode_decode_expect(buf, s->s, s->len, expect);
|
|
}
|
|
|
|
INLINE bencode_item_t *bencode_dictionary_get_expect(bencode_item_t *dict, const char *key, bencode_type_t expect) {
|
|
bencode_item_t *ret;
|
|
ret = bencode_dictionary_get(dict, key);
|
|
if (!ret || ret->type != expect)
|
|
return NULL;
|
|
return ret;
|
|
}
|
|
INLINE str *bencode_collapse_str(bencode_item_t *root, str *out) {
|
|
out->s = bencode_collapse(root, &out->len);
|
|
return out;
|
|
}
|
|
INLINE int bencode_strcmp(bencode_item_t *a, const char *b) {
|
|
int len;
|
|
if (a->type != BENCODE_STRING)
|
|
return 2;
|
|
len = strlen(b);
|
|
if (a->iov[1].iov_len < len)
|
|
return -1;
|
|
if (a->iov[1].iov_len > len)
|
|
return 1;
|
|
return memcmp(a->iov[1].iov_base, b, len);
|
|
}
|
|
INLINE int bencode_dictionary_get_strcmp(bencode_item_t *dict, const char *key, const char *s) {
|
|
bencode_item_t *i;
|
|
i = bencode_dictionary_get(dict, key);
|
|
if (!i)
|
|
return 2;
|
|
return bencode_strcmp(i, s);
|
|
}
|
|
|
|
INLINE str *bencode_get_str(bencode_item_t *in, str *out) {
|
|
if (!in || in->type != BENCODE_STRING)
|
|
return NULL;
|
|
out->s = in->iov[1].iov_base;
|
|
out->len = in->iov[1].iov_len;
|
|
return out;
|
|
}
|
|
|
|
#endif
|