buffers.c

关于tor匿名通信的源代码

{
  if (!string_len)
    return (int)buf->datalen;
  check();

  while (string_len) {
    size_t copy;
    if (!buf->tail || !CHUNK_REMAINING_CAPACITY(buf->tail))
      buf_add_chunk_with_capacity(buf, string_len, 1);

    copy = CHUNK_REMAINING_CAPACITY(buf->tail);
    if (copy > string_len)
      copy = string_len;
    memcpy(CHUNK_WRITE_PTR(buf->tail), string, copy);
    string_len -= copy;
    string += copy;
    buf->datalen += copy;
    buf->tail->datalen += copy;
  }

  check();
  tor_assert(buf->datalen < INT_MAX);
  return (int)buf->datalen;
}

/** Helper: copy the first <b>string_len</b> bytes from <b>buf</b>
 * onto <b>string</b>.
 */
static INLINE void
peek_from_buf(char *string, size_t string_len, const buf_t *buf)
{
  chunk_t *chunk;

  tor_assert(string);
  /* make sure we don't ask for too much */
  tor_assert(string_len <= buf->datalen);
  /* assert_buf_ok(buf); */

  chunk = buf->head;
  while (string_len) {
    size_t copy = string_len;
    tor_assert(chunk);
    if (chunk->datalen < copy)
      copy = chunk->datalen;
    memcpy(string, chunk->data, copy);
    string_len -= copy;
    string += copy;
    chunk = chunk->next;
  }
}

/** Remove <b>string_len</b> bytes from the front of <b>buf</b>, and store
 * them into <b>string</b>.  Return the new buffer size.  <b>string_len</b>
 * must be \<= the number of bytes on the buffer.
 */
int
fetch_from_buf(char *string, size_t string_len, buf_t *buf)
{
  /* There must be string_len bytes in buf; write them onto string,
   * then memmove buf back (that is, remove them from buf).
   *
   * Return the number of bytes still on the buffer. */

  check();
  peek_from_buf(string, string_len, buf);
  buf_remove_from_front(buf, string_len);
  check();
  tor_assert(buf->datalen < INT_MAX);
  return (int)buf->datalen;
}

/** Check <b>buf</b> for a variable-length cell according to the rules of link
 * protocol version <b>linkproto</b>.  If one is found, pull it off the buffer
 * and assign a newly allocated var_cell_t to *<b>out</b>, and return 1.
 * Return 0 if whatever is on the start of buf_t is not a variable-length
 * cell.  Return 1 and set *<b>out</b> to NULL if there seems to be the start
 * of a variable-length cell on <b>buf</b>, but the whole thing isn't there
 * yet. */
int
fetch_var_cell_from_buf(buf_t *buf, var_cell_t **out, int linkproto)
{
  char hdr[VAR_CELL_HEADER_SIZE];
  var_cell_t *result;
  uint8_t command;
  uint16_t length;
  /* If linkproto is unknown (0) or v2 (2), variable-length cells work as
   * implemented here. If it's 1, there are no variable-length cells.  Tor
   * does not support other versions right now, and so can't negotiate them.
   */
  if (linkproto == 1)
    return 0;
  check();
  *out = NULL;
  if (buf->datalen < VAR_CELL_HEADER_SIZE)
    return 0;
  peek_from_buf(hdr, sizeof(hdr), buf);

  command = *(uint8_t*)(hdr+2);
  if (!(CELL_COMMAND_IS_VAR_LENGTH(command)))
    return 0;

  length = ntohs(get_uint16(hdr+3));
  if (buf->datalen < (size_t)(VAR_CELL_HEADER_SIZE+length))
    return 1;
  result = var_cell_new(length);
  result->command = command;
  result->circ_id = ntohs(*(uint16_t*)hdr);

  buf_remove_from_front(buf, VAR_CELL_HEADER_SIZE);
  peek_from_buf(result->payload, length, buf);
  buf_remove_from_front(buf, length);
  check();

  *out = result;
  return 1;
}

/** Move up to *<b>buf_flushlen</b> bytes from <b>buf_in</b> to
 * <b>buf_out</b>, and modify *<b>buf_flushlen</b> appropriately.
 * Return the number of bytes actually copied.
 */
int
move_buf_to_buf(buf_t *buf_out, buf_t *buf_in, size_t *buf_flushlen)
{
  /* XXXX we can do way better here, but this doesn't turn up in any
   * profiles. */
  char b[4096];
  size_t cp, len;
  len = *buf_flushlen;
  if (len > buf_in->datalen)
    len = buf_in->datalen;

  cp = len; /* Remember the number of bytes we intend to copy. */
  tor_assert(cp < INT_MAX);
  while (len) {
    /* This isn't the most efficient implementation one could imagine, since
     * it does two copies instead of 1, but I kinda doubt that this will be
     * critical path. */
    size_t n = len > sizeof(b) ? sizeof(b) : len;
    fetch_from_buf(b, n, buf_in);
    write_to_buf(b, n, buf_out);
    len -= n;
  }
  *buf_flushlen -= cp;
  return (int)cp;
}

/** Internal structure: represents a position in a buffer. */
typedef struct buf_pos_t {
  const chunk_t *chunk; /**< Which chunk are we pointing to? */
  int pos;/**< Which character inside the chunk's data are we pointing to? */
  size_t chunk_pos; /**< Total length of all previous chunks. */
} buf_pos_t;

/** Initialize <b>out</b> to point to the first character of <b>buf</b>.*/
static void
buf_pos_init(const buf_t *buf, buf_pos_t *out)
{
  out->chunk = buf->head;
  out->pos = 0;
  out->chunk_pos = 0;
}

/** Advance <b>out</b> to the first appearance of <b>ch</b> at the current
 * position of <b>out</b>, or later.  Return -1 if no instances are found;
 * otherwise returns the absolute position of the character. */
static off_t
buf_find_pos_of_char(char ch, buf_pos_t *out)
{
  const chunk_t *chunk;
  int pos;
  tor_assert(out);
  if (out->chunk) {
    if (out->chunk->datalen) {
      /*XXXX020 remove this once the bug it detects is fixed. */
      if (!(out->pos < (off_t)out->chunk->datalen)) {
        log_warn(LD_BUG, "About to assert. %p, %d, %d, %p, %d.",
                 out, (int)out->pos,
                 (int)out->chunk_pos, out->chunk,
                 out->chunk?(int)out->chunk->datalen : (int)-1
                 );
      }
      tor_assert(out->pos < (off_t)out->chunk->datalen);
    } else {
      tor_assert(out->pos == 0);
    }
  }
  pos = out->pos;
  for (chunk = out->chunk; chunk; chunk = chunk->next) {
    char *cp = memchr(chunk->data+pos, ch, chunk->datalen - pos);
    if (cp) {
      out->chunk = chunk;
      tor_assert(cp - chunk->data < INT_MAX);
      out->pos = (int)(cp - chunk->data);
      return out->chunk_pos + out->pos;
    } else {
      out->chunk_pos += chunk->datalen;
      pos = 0;
    }
  }
  return -1;
}

/** Advance <b>pos</b> by a single character, if there are any more characters
 * in the buffer.  Returns 0 on sucess, -1 on failure. */
static INLINE int
buf_pos_inc(buf_pos_t *pos)
{
  ++pos->pos;
  if (pos->pos == (off_t)pos->chunk->datalen) {
    if (!pos->chunk->next)
      return -1;
    pos->chunk_pos += pos->chunk->datalen;
    pos->chunk = pos->chunk->next;
    pos->pos = 0;
  }
  return 0;
}

/** Return true iff the <b>n</b>-character string in <b>s</b> appears
 * (verbatim) at <b>pos</b>. */
static int
buf_matches_at_pos(const buf_pos_t *pos, const char *s, size_t n)
{
  buf_pos_t p;
  if (!n)
    return 1;

  memcpy(&p, pos, sizeof(p));

  while (1) {
    char ch = p.chunk->data[p.pos];
    if (ch != *s)
      return 0;
    ++s;
    /* If we're out of characters that don't match, we match.  Check this
     * _before_ we test incrementing pos, in case we're at the end of the
     * string. */
    if (--n == 0)
      return 1;
    if (buf_pos_inc(&p)<0)
      return 0;
  }
}

/** Return the first position in <b>buf</b> at which the <b>n</b>-character
 * string <b>s</b> occurs, or -1 if it does not occur. */
/*private*/ int
buf_find_string_offset(const buf_t *buf, const char *s, size_t n)
{
  buf_pos_t pos;
  buf_pos_init(buf, &pos);
  while (buf_find_pos_of_char(*s, &pos) >= 0) {
    if (buf_matches_at_pos(&pos, s, n)) {
      tor_assert(pos.chunk_pos + pos.pos < INT_MAX);
      return (int)(pos.chunk_pos + pos.pos);
    } else {
      if (buf_pos_inc(&pos)<0)
        return -1;
    }
  }
  return -1;
}

/** There is a (possibly incomplete) http statement on <b>buf</b>, of the
 * form "\%s\\r\\n\\r\\n\%s", headers, body. (body may contain nuls.)
 * If a) the headers include a Content-Length field and all bytes in
 * the body are present, or b) there's no Content-Length field and
 * all headers are present, then:
 *
 *  - strdup headers into <b>*headers_out</b>, and nul-terminate it.
 *  - memdup body into <b>*body_out</b>, and nul-terminate it.
 *  - Then remove them from <b>buf</b>, and return 1.
 *
 *  - If headers or body is NULL, discard that part of the buf.
 *  - If a headers or body doesn't fit in the arg, return -1.
 *  (We ensure that the headers or body don't exceed max len,
 *   _even if_ we're planning to discard them.)
 *  - If force_complete is true, then succeed even if not all of the
 *    content has arrived.
 *
 * Else, change nothing and return 0.
 */
int
fetch_from_buf_http(buf_t *buf,
                    char **headers_out, size_t max_headerlen,
                    char **body_out, size_t *body_used, size_t max_bodylen,
                    int force_complete)
{
  char *headers, *p;
  size_t headerlen, bodylen, contentlen;
  int crlf_offset;

  check();
  if (!buf->head)
    return 0;

  crlf_offset = buf_find_string_offset(buf, "\r\n\r\n", 4);
  if (crlf_offset > (int)max_headerlen ||
      (crlf_offset < 0 && buf->datalen > max_headerlen)) {
    log_debug(LD_HTTP,"headers too long.");
    return -1;
  } else if (crlf_offset < 0) {
    log_debug(LD_HTTP,"headers not all here yet.");
    return 0;
  }
  /* Okay, we have a full header.  Make sure it all appears in the first
   * chunk. */
  if ((int)buf->head->datalen < crlf_offset + 4)
    buf_pullup(buf, crlf_offset+4, 0);
  headerlen = crlf_offset + 4;

  headers = buf->head->data;
  bodylen = buf->datalen - headerlen;
  log_debug(LD_HTTP,"headerlen %d, bodylen %d.", (int)headerlen, (int)bodylen);

  if (max_headerlen <= headerlen) {
    log_warn(LD_HTTP,"headerlen %d larger than %d. Failing.",
             (int)headerlen, (int)max_headerlen-1);
    return -1;
  }
  if (max_bodylen <= bodylen) {
    log_warn(LD_HTTP,"bodylen %d larger than %d. Failing.",
             (int)bodylen, (int)max_bodylen-1);
    return -1;
  }

#define CONTENT_LENGTH "\r\nContent-Length: "
  p = (char*) tor_memstr(headers, headerlen, CONTENT_LENGTH);
  if (p) {
    int i;
    i = atoi(p+strlen(CONTENT_LENGTH));
    if (i < 0) {
      log_warn(LD_PROTOCOL, "Content-Length is less than zero; it looks like "
               "someone is trying to crash us.");
      return -1;
    }
    contentlen = i;
    /* if content-length is malformed, then our body length is 0. fine. */
    log_debug(LD_HTTP,"Got a contentlen of %d.",(int)contentlen);
    if (bodylen < contentlen) {
      if (!force_complete) {
        log_debug(LD_HTTP,"body not all here yet.");
        return 0; /* not all there yet */
      }
    }
    if (bodylen > contentlen) {
      bodylen = contentlen;
      log_debug(LD_HTTP,"bodylen reduced to %d.",(int)bodylen);
    }
  }
  /* all happy. copy into the appropriate places, and return 1 */
  if (headers_out) {
    *headers_out = tor_malloc(headerlen+1);
    fetch_from_buf(*headers_out, headerlen, buf);
    (*headers_out)[headerlen] = 0; /* nul terminate it */
  }
  if (body_out) {
    tor_assert(body_used);
    *body_used = bodylen;
    *body_out = tor_malloc(bodylen+1);
    fetch_from_buf(*body_out, bodylen, buf);
    (*body_out)[bodylen] = 0; /* nul terminate it */
  }
  check();
  return 1;
}

/** There is a (possibly incomplete) socks handshake on <b>buf</b>, of one
 * of the forms
 *  - socks4: "socksheader username\\0"
 *  - socks4a: "socksheader username\\0 destaddr\\0"
 *  - socks5 phase one: "version #methods methods"
 *  - socks5 phase two: "version command 0 addresstype..."
 * If it's a complete and valid handshake, and destaddr fits in
 *   MAX_SOCKS_ADDR_LEN bytes, then pull the handshake off the buf,
 *   assign to <b>req</b>, and return 1.
 *
 * If it's invalid or too big, return -1.
 *
 * Else it's not all there yet, leave buf alone and return 0.
 *
 * If you want to specify the socks reply, write it into <b>req->reply</b>
 *   and set <b>req->replylen</b>, else leave <b>req->replylen</b> alone.
 *
 * If <b>log_sockstype</b> is non-zero, then do a notice-level log of whether
 * the connection is possibly leaking DNS requests locally or not.
 *
 * If <b>safe_socks</b> is true, then reject unsafe socks protocols.
 *
 * If returning 0 or -1, <b>req->address</b> and <b>req->port</b> are
 * undefined.
 */
int
fetch_from_buf_socks(buf_t *buf, socks_request_t *req,
                     int log_sockstype, int safe_socks)
{
  unsigned int len;
  char tmpbuf[INET_NTOA_BUF_LEN];
  uint32_t destip;
  uint8_t socksver;
  enum {socks4, socks4a} socks4_prot = socks4a;
  char *next, *startaddr;
  struct in_addr in;

  /* If the user connects with socks4 or the wrong variant of socks5,
   * then log a warning to let him know that it might be unwise. */
  static int have_warned_about_unsafe_socks = 0;

  if (buf->datalen < 2) /* version and another byte */
    return 0;

  buf_pullup(buf, 128, 0);
  tor_assert(buf->head && buf->head->datalen >= 2);

  socksver = *buf->head->data;

  switch (socksver) { /* which version of socks? */

    case 5: /* socks5 */

      if (req->socks_version != 5) { /* we need to negotiate a method */
        unsigned char nummethods = (unsigned char)*(buf->head->data+1);
        tor_assert(!req->socks_version);
        if (buf->datalen < 2u+nummethods)
          return 0;
        buf_pullup(buf, 2u+nummethods, 0);
        if (!nummethods || !memchr(buf->head->data+2, 0, nummethods)) {
          log_warn(LD_APP,
                   "socks5: offered methods don't include 'no auth'. "