Reduce number of places knowing about filename-construction for mbox file-for-scanning

[exim.git] / src / src / malware.c

/*************************************************
*     Exim - an Internet mail transport agent    *
*************************************************/

/* Copyright (c) Tom Kistner <tom@duncanthrax.net> 2003 - 2015
 * License: GPL
 * Copyright (c) The Exim Maintainers 2016
 */

/* Code for calling virus (malware) scanners. Called from acl.c. */

#include "exim.h"
#ifdef WITH_CONTENT_SCAN

typedef enum {M_FPROTD, M_DRWEB, M_AVES, M_FSEC, M_KAVD, M_CMDL,
                M_SOPHIE, M_CLAMD, M_SOCK, M_MKSD, M_AVAST, M_FPROT6D} scanner_t;
typedef enum {MC_NONE, MC_TCP, MC_UNIX, MC_STRM} contype_t;
static struct scan
{
  scanner_t     scancode;
  const uschar * name;
  const uschar * options_default;
  contype_t     conn;
} m_scans[] =
{
  { M_FPROTD,   US"f-protd",    US"localhost 10200-10204",            MC_TCP },
  { M_DRWEB,    US"drweb",      US"/usr/local/drweb/run/drwebd.sock", MC_STRM },
  { M_AVES,     US"aveserver",  US"/var/run/aveserver",               MC_UNIX },
  { M_FSEC,     US"fsecure",    US"/var/run/.fsav",                   MC_UNIX },
  { M_KAVD,     US"kavdaemon",  US"/var/run/AvpCtl",                  MC_UNIX },
  { M_CMDL,     US"cmdline",    NULL,                                 MC_NONE },
  { M_SOPHIE,   US"sophie",     US"/var/run/sophie",                  MC_UNIX },
  { M_CLAMD,    US"clamd",      US"/tmp/clamd",                       MC_NONE },
  { M_SOCK,     US"sock",       US"/tmp/malware.sock",                MC_STRM },
  { M_MKSD,     US"mksd",       NULL,                                 MC_NONE },
  { M_AVAST,    US"avast",      US"/var/run/avast/scan.sock",         MC_STRM },
  { M_FPROT6D,  US"f-prot6d",   US"localhost 10200",                  MC_TCP },
  { -1,         NULL,           NULL, MC_NONE }         /* end-marker */
};

/* The maximum number of clamd servers that are supported in the configuration */
#define MAX_CLAMD_SERVERS 32
#define MAX_CLAMD_SERVERS_S "32"

typedef struct clamd_address {
  uschar * hostspec;
  unsigned tcp_port;
  unsigned retry;
} clamd_address;

#ifndef nelements
# define nelements(arr) (sizeof(arr) / sizeof(arr[0]))
#endif


#define MALWARE_TIMEOUT 120     /* default timeout, seconds */


#define DRWEBD_SCAN_CMD             (1)     /* scan file, buffer or diskfile */
#define DRWEBD_RETURN_VIRUSES       (1<<0)   /* ask daemon return to us viruses names from report */
#define DRWEBD_IS_MAIL              (1<<19)  /* say to daemon that format is "archive MAIL" */

#define DERR_READ_ERR               (1<<0)   /* read error */
#define DERR_NOMEMORY               (1<<2)   /* no memory */
#define DERR_TIMEOUT                (1<<9)   /* scan timeout has run out */
#define DERR_BAD_CALL               (1<<15)  /* wrong command */


static const uschar * malware_regex_default = US ".+";
static const pcre * malware_default_re = NULL;

static const uschar * drweb_re_str = US "infected\\swith\\s*(.+?)$";
static const pcre * drweb_re = NULL;

static const uschar * fsec_re_str = US "\\S{0,5}INFECTED\\t[^\\t]*\\t([^\\t]+)\\t\\S*$";
static const pcre * fsec_re = NULL;

static const uschar * kav_re_sus_str = US "suspicion:\\s*(.+?)\\s*$";
static const uschar * kav_re_inf_str = US "infected:\\s*(.+?)\\s*$";
static const pcre * kav_re_sus = NULL;
static const pcre * kav_re_inf = NULL;

static const uschar * ava_re_clean_str = US "(?!\\\\)\\t\\[\\+\\]";
static const uschar * ava_re_virus_str = US "(?!\\\\)\\t\\[L\\]\\d\\.\\d\\t\\d\\s(.*)";
static const pcre * ava_re_clean = NULL;
static const pcre * ava_re_virus = NULL;

static const uschar * fprot6d_re_error_str = US "^\\d+\\s<(.+?)>$";
static const uschar * fprot6d_re_virus_str = US "^\\d+\\s<infected:\\s+(.+?)>\\s+.+$";
static const pcre * fprot6d_re_error = NULL;
static const pcre * fprot6d_re_virus = NULL;



/******************************************************************************/

/* Routine to check whether a system is big- or little-endian.
   Ripped from http://www.faqs.org/faqs/graphics/fileformats-faq/part4/section-7.html
   Needed for proper kavdaemon implementation. Sigh. */
#define BIG_MY_ENDIAN      0
#define LITTLE_MY_ENDIAN   1
static int test_byte_order(void);
static inline int
test_byte_order()
{
  short int word = 0x0001;
  char *byte = (char *) &word;
  return(byte[0] ? LITTLE_MY_ENDIAN : BIG_MY_ENDIAN);
}

BOOL malware_ok = FALSE;

/* Gross hacks for the -bmalware option; perhaps we should just create
the scan directory normally for that case, but look into rigging up the
needed header variables if not already set on the command-line? */
extern int spool_mbox_ok;
extern uschar spooled_message_id[MESSAGE_ID_LENGTH+1];



static inline int
malware_errlog_defer(const uschar * str)
{
log_write(0, LOG_MAIN|LOG_PANIC, "malware acl condition: %s", str);
return DEFER;
}

static int
m_errlog_defer(struct scan * scanent, const uschar * hostport,
  const uschar * str)
{
return malware_errlog_defer(string_sprintf("%s %s : %s",
  scanent->name, hostport ? hostport : CUS"", str));
}
static int
m_errlog_defer_3(struct scan * scanent, const uschar * hostport,
  const uschar * str, int fd_to_close)
{
(void) close(fd_to_close);
return m_errlog_defer(scanent, hostport, str);
}

/*************************************************/

/* Only used by the Clamav code, which is working from a list of servers and
uses the returned in_addr to get a second connection to the same system.
*/
static inline int
m_tcpsocket(const uschar * hostname, unsigned int port,
        host_item * host, uschar ** errstr)
{
return ip_connectedsocket(SOCK_STREAM, hostname, port, port, 5, host, errstr);
}

static int
m_sock_send(int sock, uschar * buf, int cnt, uschar ** errstr)
{
if (send(sock, buf, cnt, 0) < 0)
  {
  int err = errno;
  (void)close(sock);
  *errstr = string_sprintf("unable to send to socket (%s): %s",
         buf, strerror(err));
  return -1;
  }
return sock;
}

static const pcre *
m_pcre_compile(const uschar * re, uschar ** errstr)
{
const uschar * rerror;
int roffset;
const pcre * cre;

cre = pcre_compile(CS re, PCRE_COPT, (const char **)&rerror, &roffset, NULL);
if (!cre)
  *errstr= string_sprintf("regular expression error in '%s': %s at offset %d",
      re, rerror, roffset);
return cre;
}

uschar *
m_pcre_exec(const pcre * cre, uschar * text)
{
int ovector[10*3];
int i = pcre_exec(cre, NULL, CS text, Ustrlen(text), 0, 0,
              ovector, nelements(ovector));
uschar * substr = NULL;
if (i >= 2)                             /* Got it */
  pcre_get_substring(CS text, ovector, i, 1, (const char **) &substr);
return substr;
}

static const pcre *
m_pcre_nextinlist(const uschar ** list, int * sep,
 char * listerr, uschar ** errstr)
{
const uschar * list_ele;
const pcre * cre = NULL;

if (!(list_ele = string_nextinlist(list, sep, NULL, 0)))
  *errstr = US listerr;
else
  cre = m_pcre_compile(CUS list_ele, errstr);
return cre;
}

/*
 Simple though inefficient wrapper for reading a line.  Drop CRs and the
 trailing newline. Can return early on buffer full. Null-terminate.
 Apply initial timeout if no data ready.

 Return: number of chars - zero for an empty line
         -1 on EOF
         -2 on timeout or error
*/
static int
recv_line(int fd, uschar * buffer, int bsize, int tmo)
{
uschar * p = buffer;
ssize_t rcv;
BOOL ok = FALSE;

if (!fd_ready(fd, tmo-time(NULL)))
  return -2;

/*XXX tmo handling assumes we always get a whole line */
/* read until \n */
errno = 0;
while ((rcv = read(fd, p, 1)) > 0)
  {
  ok = TRUE;
  if (p-buffer > bsize-2) break;
  if (*p == '\n') break;
  if (*p != '\r') p++;
  }
if (!ok)
  {
  DEBUG(D_acl) debug_printf_indent("Malware scan: read %s (%s)\n",
                rcv==0 ? "EOF" : "error", strerror(errno));
  return rcv==0 ? -1 : -2;
  }
*p = '\0';

DEBUG(D_acl) debug_printf_indent("Malware scan: read '%s'\n", buffer);
return p - buffer;
}

/* return TRUE iff size as requested */
static BOOL
recv_len(int sock, void * buf, int size, int tmo)
{
return fd_ready(sock, tmo-time(NULL))
  ? recv(sock, buf, size, 0) == size
  : FALSE;
}



/* ============= private routines for the "mksd" scanner type ============== */

#include <sys/uio.h>

static inline int
mksd_writev (int sock, struct iovec * iov, int iovcnt)
{
int i;

for (;;)
  {
  do
    i = writev (sock, iov, iovcnt);
  while (i < 0 && errno == EINTR);
  if (i <= 0)
    {
    (void) malware_errlog_defer(
            US"unable to write to mksd UNIX socket (/var/run/mksd/socket)");
    return -1;
    }
  for (;;)      /* check for short write */
    if (i >= iov->iov_len)
      {
      if (--iovcnt == 0)
        return 0;
      i -= iov->iov_len;
      iov++;
      }
    else
      {
      iov->iov_len -= i;
      iov->iov_base = CS iov->iov_base + i;
      break;
      }
  }
}

static inline int
mksd_read_lines (int sock, uschar *av_buffer, int av_buffer_size, int tmo)
{
int offset = 0;
int i;

do
  {
  i = ip_recv(sock, av_buffer+offset, av_buffer_size-offset, tmo-time(NULL));
  if (i <= 0)
    {
    (void) malware_errlog_defer(US"unable to read from mksd UNIX socket (/var/run/mksd/socket)");
    return -1;
    }

  offset += i;
  /* offset == av_buffer_size -> buffer full */
  if (offset == av_buffer_size)
    {
    (void) malware_errlog_defer(US"malformed reply received from mksd");
    return -1;
    }
  } while (av_buffer[offset-1] != '\n');

av_buffer[offset] = '\0';
return offset;
}

static inline int
mksd_parse_line(struct scan * scanent, char * line)
{
char *p;

switch (*line)
  {
  case 'O': /* OK */
    return OK;

  case 'E':
  case 'A': /* ERR */
    if ((p = strchr (line, '\n')) != NULL)
      *p = '\0';
    return m_errlog_defer(scanent, NULL,
      string_sprintf("scanner failed: %s", line));

  default: /* VIR */
    if ((p = strchr (line, '\n')) != NULL)
      {
      *p = '\0';
      if (  p-line > 5
         && line[3] == ' '
         && (p = strchr(line+4, ' ')) != NULL
         && p-line > 4
         )
        {
        *p = '\0';
        malware_name = string_copy(US line+4);
        return OK;
        }
      }
    return m_errlog_defer(scanent, NULL,
      string_sprintf("malformed reply received: %s", line));
  }
}

static int
mksd_scan_packed(struct scan * scanent, int sock, const uschar * scan_filename,
  int tmo)
{
struct iovec iov[3];
const char *cmd = "MSQ\n";
uschar av_buffer[1024];

iov[0].iov_base = (void *) cmd;
iov[0].iov_len = 3;
iov[1].iov_base = (void *) scan_filename;
iov[1].iov_len = Ustrlen(scan_filename);
iov[2].iov_base = (void *) (cmd + 3);
iov[2].iov_len = 1;

if (mksd_writev (sock, iov, 3) < 0)
  return DEFER;

if (mksd_read_lines (sock, av_buffer, sizeof (av_buffer), tmo) < 0)
  return DEFER;

return mksd_parse_line (scanent, CS av_buffer);
}


static int
clamd_option(clamd_address * cd, const uschar * optstr, int * subsep)
{
uschar * s;

cd->retry = 0;
while ((s = string_nextinlist(&optstr, subsep, NULL, 0)))
  if (Ustrncmp(s, "retry=", 6) == 0)
    {
    int sec = readconf_readtime((s += 6), '\0', FALSE);
    if (sec < 0)
      return FAIL;
    cd->retry = sec;
    }
  else
    return FAIL;
return OK;
}

/*************************************************
*          Scan content for malware              *
*************************************************/

/* This is an internal interface for scanning an email; the normal interface
is via malware(), or there's malware_in_file() used for testing/debugging.

Arguments:
  malware_re    match condition for "malware="
  scan_filename  the file holding the email to be scanned, if we're faking
                this up for the -bmalware test, else NULL
  timeout       if nonzero, non-default timeoutl

Returns:        Exim message processing code (OK, FAIL, DEFER, ...)
                where true means malware was found (condition applies)
*/
static int
malware_internal(const uschar * malware_re, const uschar * scan_filename,
  int timeout)
{
int sep = 0;
const uschar *av_scanner_work = av_scanner;
uschar *scanner_name;
unsigned long mbox_size;
FILE *mbox_file;
const pcre *re;
uschar * errstr;
struct scan * scanent;
const uschar * scanner_options;
int sock = -1;
time_t tmo;
uschar * eml_filename, * eml_dir;

if (!malware_re)
  return FAIL;          /* empty means "don't match anything" */

/* Ensure the eml mbox file is spooled up */

if (!(mbox_file = spool_mbox(&mbox_size, scan_filename, &eml_filename)))
  return malware_errlog_defer(US"error while creating mbox spool file");

/* None of our current scanners need the mbox file as a stream (they use
the name), so we can close it right away.  Get the directory too. */

(void) fclose(mbox_file);
eml_dir = string_copyn(eml_filename, Ustrrchr(eml_filename, '/') - eml_filename);

/* parse 1st option */
if (strcmpic(malware_re, US"false") == 0  ||  Ustrcmp(malware_re,"0") == 0)
  return FAIL;          /* explicitly no matching */

/* special cases (match anything except empty) */
if (  strcmpic(malware_re,US"true") == 0
   || Ustrcmp(malware_re,"*") == 0
   || Ustrcmp(malware_re,"1") == 0
   )
  {
  if (  !malware_default_re
     && !(malware_default_re = m_pcre_compile(malware_regex_default, &errstr)))
    return malware_errlog_defer(errstr);
  malware_re = malware_regex_default;
  re = malware_default_re;
  }

/* compile the regex, see if it works */
else if (!(re = m_pcre_compile(malware_re, &errstr)))
  return malware_errlog_defer(errstr);

/* Reset sep that is set by previous string_nextinlist() call */
sep = 0;

/* if av_scanner starts with a dollar, expand it first */
if (*av_scanner == '$')
  {
  if (!(av_scanner_work = expand_string(av_scanner)))
    return malware_errlog_defer(
         string_sprintf("av_scanner starts with $, but expansion failed: %s",
         expand_string_message));

  DEBUG(D_acl)
    debug_printf_indent("Expanded av_scanner global: %s\n", av_scanner_work);
  /* disable result caching in this case */
  malware_name = NULL;
  malware_ok = FALSE;
  }

/* Do not scan twice (unless av_scanner is dynamic). */
if (!malware_ok)
  {
  /* find the scanner type from the av_scanner option */
  if (!(scanner_name = string_nextinlist(&av_scanner_work, &sep, NULL, 0)))
    return malware_errlog_defer(US"av_scanner configuration variable is empty");
  if (!timeout) timeout = MALWARE_TIMEOUT;
  tmo = time(NULL) + timeout;

  for (scanent = m_scans; ; scanent++)
    {
    if (!scanent->name)
      return malware_errlog_defer(string_sprintf("unknown scanner type '%s'",
        scanner_name));
    if (strcmpic(scanner_name, US scanent->name) != 0)
      continue;
    if (!(scanner_options = string_nextinlist(&av_scanner_work, &sep, NULL, 0)))
      scanner_options = scanent->options_default;
    if (scanent->conn == MC_NONE)
      break;
    switch(scanent->conn)
    {
    case MC_TCP:  sock = ip_tcpsocket(scanner_options, &errstr, 5);     break;
    case MC_UNIX: sock = ip_unixsocket(scanner_options, &errstr);       break;
    case MC_STRM: sock = ip_streamsocket(scanner_options, &errstr, 5);  break;
    default: /* compiler quietening */ break;
    }
    if (sock < 0)
      return m_errlog_defer(scanent, CUS callout_address, errstr);
    break;
  }
  DEBUG(D_acl) debug_printf_indent("Malware scan: %s tmo %s\n", scanner_name, readconf_printtime(timeout));

  switch (scanent->scancode)
    {
    case M_FPROTD: /* "f-protd" scanner type -------------------------------- */
      {
      uschar *fp_scan_option;
      unsigned int detected=0, par_count=0;
      uschar * scanrequest;
      uschar buf[32768], *strhelper, *strhelper2;
      uschar * malware_name_internal = NULL;
      int len;

      scanrequest = string_sprintf("GET %s", eml_filename);

      while ((fp_scan_option = string_nextinlist(&av_scanner_work, &sep,
                            NULL, 0)))
        {
        scanrequest = string_sprintf("%s%s%s", scanrequest,
                                  par_count ? "%20" : "?", fp_scan_option);
        par_count++;
        }
      scanrequest = string_sprintf("%s HTTP/1.0\r\n\r\n", scanrequest);
      DEBUG(D_acl) debug_printf_indent("Malware scan: issuing %s: %s\n",
        scanner_name, scanrequest);

      /* send scan request */
      if (m_sock_send(sock, scanrequest, Ustrlen(scanrequest)+1, &errstr) < 0)
        return m_errlog_defer(scanent, CUS callout_address, errstr);

      while ((len = recv_line(sock, buf, sizeof(buf), tmo)) >= 0)
        if (len > 0)
          {
          if (Ustrstr(buf, US"<detected type=\"") != NULL)
            detected = 1;
          else if (detected && (strhelper = Ustrstr(buf, US"<name>")))
            {
            if ((strhelper2 = Ustrstr(buf, US"</name>")) != NULL)
              {
              *strhelper2 = '\0';
              malware_name_internal = string_copy(strhelper+6);
              }
            }
          else if (Ustrstr(buf, US"<summary code=\""))
            {
            malware_name = Ustrstr(buf, US"<summary code=\"11\">")
                ? malware_name_internal : NULL;
            break;
            }
          }
      if (len < -1)
        {
        (void)close(sock);
        return DEFER;
        }
      break;
      } /* f-protd */

    case M_DRWEB: /* "drweb" scanner type ----------------------------------- */
  /* v0.1 - added support for tcp sockets          */
  /* v0.0 - initial release -- support for unix sockets      */
      {
      int result;
      off_t fsize;
      unsigned int fsize_uint;
      uschar * tmpbuf, *drweb_fbuf;
      int drweb_rc, drweb_cmd, drweb_flags = 0x0000, drweb_fd,
          drweb_vnum, drweb_slen, drweb_fin = 0x0000;

      /* prepare variables */
      drweb_cmd = htonl(DRWEBD_SCAN_CMD);
      drweb_flags = htonl(DRWEBD_RETURN_VIRUSES | DRWEBD_IS_MAIL);

      if (*scanner_options != '/')
        {
        /* calc file size */
        if ((drweb_fd = open(CCS eml_filename, O_RDONLY)) == -1)
          return m_errlog_defer_3(scanent, NULL,
            string_sprintf("can't open spool file %s: %s",
              eml_filename, strerror(errno)),
            sock);

        if ((fsize = lseek(drweb_fd, 0, SEEK_END)) == -1)
          {
          int err = errno;
          (void)close(drweb_fd);
          return m_errlog_defer_3(scanent, NULL,
            string_sprintf("can't seek spool file %s: %s",
              eml_filename, strerror(err)),
            sock);
          }
        fsize_uint = (unsigned int) fsize;
        if ((off_t)fsize_uint != fsize)
          {
          (void)close(drweb_fd);
          return m_errlog_defer_3(scanent, NULL,
            string_sprintf("seeking spool file %s, size overflow",
              eml_filename),
            sock);
          }
        drweb_slen = htonl(fsize);
        lseek(drweb_fd, 0, SEEK_SET);

        DEBUG(D_acl) debug_printf_indent("Malware scan: issuing %s remote scan [%s]\n",
            scanner_name, scanner_options);

        /* send scan request */
        if ((send(sock, &drweb_cmd, sizeof(drweb_cmd), 0) < 0) ||
            (send(sock, &drweb_flags, sizeof(drweb_flags), 0) < 0) ||
            (send(sock, &drweb_fin, sizeof(drweb_fin), 0) < 0) ||
            (send(sock, &drweb_slen, sizeof(drweb_slen), 0) < 0))
          {
          (void)close(drweb_fd);
          return m_errlog_defer_3(scanent, CUS callout_address, string_sprintf(
            "unable to send commands to socket (%s)", scanner_options),
            sock);
          }

        if (!(drweb_fbuf = US malloc(fsize_uint)))
          {
          (void)close(drweb_fd);
          return m_errlog_defer_3(scanent, NULL,
            string_sprintf("unable to allocate memory %u for file (%s)",
              fsize_uint, eml_filename),
            sock);
          }

        if ((result = read (drweb_fd, drweb_fbuf, fsize)) == -1)
          {
          int err = errno;
          (void)close(drweb_fd);
          free(drweb_fbuf);
          return m_errlog_defer_3(scanent, NULL,
            string_sprintf("can't read spool file %s: %s",
              eml_filename, strerror(err)),
            sock);
          }
        (void)close(drweb_fd);

        /* send file body to socket */
        if (send(sock, drweb_fbuf, fsize, 0) < 0)
          {
          free(drweb_fbuf);
          return m_errlog_defer_3(scanent, CUS callout_address, string_sprintf(
            "unable to send file body to socket (%s)", scanner_options),
            sock);
          }
        }
      else
        {
        drweb_slen = htonl(Ustrlen(eml_filename));

        DEBUG(D_acl) debug_printf_indent("Malware scan: issuing %s local scan [%s]\n",
            scanner_name, scanner_options);

        /* send scan request */
        if ((send(sock, &drweb_cmd, sizeof(drweb_cmd), 0) < 0) ||
            (send(sock, &drweb_flags, sizeof(drweb_flags), 0) < 0) ||
            (send(sock, &drweb_slen, sizeof(drweb_slen), 0) < 0) ||
            (send(sock, eml_filename, Ustrlen(eml_filename), 0) < 0) ||
            (send(sock, &drweb_fin, sizeof(drweb_fin), 0) < 0))
          return m_errlog_defer_3(scanent, CUS callout_address, string_sprintf(
            "unable to send commands to socket (%s)", scanner_options),
            sock);
        }

      /* wait for result */
      if (!recv_len(sock, &drweb_rc, sizeof(drweb_rc), tmo))
        return m_errlog_defer_3(scanent, CUS callout_address,
                    US"unable to read return code", sock);
      drweb_rc = ntohl(drweb_rc);

      if (!recv_len(sock, &drweb_vnum, sizeof(drweb_vnum), tmo))
        return m_errlog_defer_3(scanent, CUS callout_address,
                            US"unable to read the number of viruses", sock);
      drweb_vnum = ntohl(drweb_vnum);

      /* "virus(es) found" if virus number is > 0 */
      if (drweb_vnum)
        {
        int i;

        /* setup default virus name */
        malware_name = US"unknown";

        /* set up match regex */
        if (!drweb_re)
          drweb_re = m_pcre_compile(drweb_re_str, &errstr);

        /* read and concatenate virus names into one string */
        for (i = 0; i < drweb_vnum; i++)
          {
          int size = 0, off = 0, ovector[10*3];
          /* read the size of report */
          if (!recv_len(sock, &drweb_slen, sizeof(drweb_slen), tmo))
            return m_errlog_defer_3(scanent, CUS callout_address,
                              US"cannot read report size", sock);
          drweb_slen = ntohl(drweb_slen);
          tmpbuf = store_get(drweb_slen);

          /* read report body */
          if (!recv_len(sock, tmpbuf, drweb_slen, tmo))
            return m_errlog_defer_3(scanent, CUS callout_address,
                              US"cannot read report string", sock);
          tmpbuf[drweb_slen] = '\0';

          /* try matcher on the line, grab substring */
          result = pcre_exec(drweb_re, NULL, CS tmpbuf, Ustrlen(tmpbuf), 0, 0,
                                  ovector, nelements(ovector));
          if (result >= 2)
            {
            const char * pre_malware_nb;

            pcre_get_substring(CS tmpbuf, ovector, result, 1, &pre_malware_nb);

            if (i==0)   /* the first name we just copy to malware_name */
              malware_name = string_append(NULL, &size, &off,
                                          1, pre_malware_nb);

            else        /* concatenate each new virus name to previous */
              malware_name = string_append(malware_name, &size, &off,
                                          2, "/", pre_malware_nb);

            pcre_free_substring(pre_malware_nb);
            }
          }
        }
      else
        {
        const char *drweb_s = NULL;

        if (drweb_rc & DERR_READ_ERR) drweb_s = "read error";
        if (drweb_rc & DERR_NOMEMORY) drweb_s = "no memory";
        if (drweb_rc & DERR_TIMEOUT)  drweb_s = "timeout";
        if (drweb_rc & DERR_BAD_CALL) drweb_s = "wrong command";
        /* retcodes DERR_SYMLINK, DERR_NO_REGFILE, DERR_SKIPPED.
         * DERR_TOO_BIG, DERR_TOO_COMPRESSED, DERR_SPAM,
         * DERR_CRC_ERROR, DERR_READSOCKET, DERR_WRITE_ERR
         * and others are ignored */
        if (drweb_s)
          return m_errlog_defer_3(scanent, CUS callout_address,
            string_sprintf("drweb daemon retcode 0x%x (%s)", drweb_rc, drweb_s),
            sock);

        /* no virus found */
        malware_name = NULL;
        }
      break;
      } /* drweb */

    case M_AVES: /* "aveserver" scanner type -------------------------------- */
      {
      uschar buf[32768];
      int result;

      /* read aveserver's greeting and see if it is ready (2xx greeting) */
      buf[0] = 0;
      recv_line(sock, buf, sizeof(buf), tmo);

      if (buf[0] != '2')                /* aveserver is having problems */
        return m_errlog_defer_3(scanent, CUS callout_address,
          string_sprintf("unavailable (Responded: %s).",
                          ((buf[0] != 0) ? buf : (uschar *)"nothing") ),
          sock);

      /* prepare our command */
      (void)string_format(buf, sizeof(buf), "SCAN bPQRSTUW %s\r\n",
                                                eml_filename);

      /* and send it */
      DEBUG(D_acl) debug_printf_indent("Malware scan: issuing %s %s\n",
        scanner_name, buf);
      if (m_sock_send(sock, buf, Ustrlen(buf), &errstr) < 0)
        return m_errlog_defer(scanent, CUS callout_address, errstr);

      malware_name = NULL;
      result = 0;
      /* read response lines, find malware name and final response */
      while (recv_line(sock, buf, sizeof(buf), tmo) > 0)
        {
        if (buf[0] == '2')
          break;
        if (buf[0] == '5')              /* aveserver is having problems */
          {
          result = m_errlog_defer(scanent, CUS callout_address,
             string_sprintf("unable to scan file %s (Responded: %s).",
                             eml_filename, buf));
          break;
          }
        if (Ustrncmp(buf,"322",3) == 0)
          {
          uschar *p = Ustrchr(&buf[4], ' ');
          *p = '\0';
          malware_name = string_copy(&buf[4]);
          }
        }

      if (m_sock_send(sock, US"quit\r\n", 6, &errstr) < 0)
        return m_errlog_defer(scanent, CUS callout_address, errstr);

      /* read aveserver's greeting and see if it is ready (2xx greeting) */
      buf[0] = 0;
      recv_line(sock, buf, sizeof(buf), tmo);

      if (buf[0] != '2')                /* aveserver is having problems */
        return m_errlog_defer_3(scanent, CUS callout_address,
          string_sprintf("unable to quit dialogue (Responded: %s).",
                        ((buf[0] != 0) ? buf : (uschar *)"nothing") ),
          sock);

      if (result == DEFER)
        {
        (void)close(sock);
        return DEFER;
        }
      break;
      } /* aveserver */

    case M_FSEC: /* "fsecure" scanner type ---------------------------------- */
      {
      int i, j, bread = 0;
      uschar * file_name;
      uschar av_buffer[1024];
      static uschar *cmdopt[] = { US"CONFIGURE\tARCHIVE\t1\n",
                                      US"CONFIGURE\tTIMEOUT\t0\n",
                                      US"CONFIGURE\tMAXARCH\t5\n",
                                      US"CONFIGURE\tMIME\t1\n" };

      malware_name = NULL;

      DEBUG(D_acl) debug_printf_indent("Malware scan: issuing %s scan [%s]\n",
          scanner_name, scanner_options);
      /* pass options */
      memset(av_buffer, 0, sizeof(av_buffer));
      for (i = 0; i != nelements(cmdopt); i++)
        {

        if (m_sock_send(sock, cmdopt[i], Ustrlen(cmdopt[i]), &errstr) < 0)
          return m_errlog_defer(scanent, CUS callout_address, errstr);

        bread = ip_recv(sock, av_buffer, sizeof(av_buffer), tmo-time(NULL));
        if (bread > 0) av_buffer[bread]='\0';
        if (bread < 0)
          return m_errlog_defer_3(scanent, CUS callout_address,
            string_sprintf("unable to read answer %d (%s)", i, strerror(errno)),
            sock);
        for (j = 0; j < bread; j++)
          if (av_buffer[j] == '\r' || av_buffer[j] == '\n')
            av_buffer[j] ='@';
        }

      /* pass the mailfile to fsecure */
      file_name = string_sprintf("SCAN\t%s\n", eml_filename);

      if (m_sock_send(sock, file_name, Ustrlen(file_name), &errstr) < 0)
        return m_errlog_defer(scanent, CUS callout_address, errstr);

      /* set up match */
      /* todo also SUSPICION\t */
      if (!fsec_re)
        fsec_re = m_pcre_compile(fsec_re_str, &errstr);

      /* read report, linewise. Apply a timeout as the Fsecure daemon
      sometimes wants an answer to "PING" but they won't tell us what */
        {
        uschar * p = av_buffer;
        uschar * q;

        for (;;)
          {
          errno = ETIMEDOUT;
          i =  av_buffer+sizeof(av_buffer)-p;
          if ((bread= ip_recv(sock, p, i-1, tmo-time(NULL))) < 0)
            return m_errlog_defer_3(scanent, CUS callout_address,
              string_sprintf("unable to read result (%s)", strerror(errno)),
              sock);

          for (p[bread] = '\0'; (q = Ustrchr(p, '\n')); p = q+1)
            {
            *q = '\0';

            /* Really search for virus again? */
            if (!malware_name)
              /* try matcher on the line, grab substring */
              malware_name = m_pcre_exec(fsec_re, p);

            if (Ustrstr(p, "OK\tScan ok."))
              goto fsec_found;
            }

          /* copy down the trailing partial line then read another chunk */
          i =  av_buffer+sizeof(av_buffer)-p;
          memmove(av_buffer, p, i);
          p = av_buffer+i;
          }
        }

      fsec_found:
        break;
      } /* fsecure */

    case M_KAVD: /* "kavdaemon" scanner type -------------------------------- */
      {
      time_t t;
      uschar tmpbuf[1024];
      uschar * scanrequest;
      int kav_rc;
      unsigned long kav_reportlen;
      int bread;
      const pcre *kav_re;
      uschar *p;

      /* get current date and time, build scan request */
      time(&t);
      /* pdp note: before the eml_filename parameter, this scanned the
      directory; not finding documentation, so we'll strip off the directory.
      The side-effect is that the test framework scanning may end up in
      scanning more than was requested, but for the normal interface, this is
      fine. */

      strftime(CS tmpbuf, sizeof(tmpbuf), "%d %b %H:%M:%S", localtime(&t));
      scanrequest = string_sprintf("<0>%s:%s", CS tmpbuf, eml_filename);
      p = Ustrrchr(scanrequest, '/');
      if (p)
        *p = '\0';

      DEBUG(D_acl) debug_printf_indent("Malware scan: issuing %s scan [%s]\n",
          scanner_name, scanner_options);

      /* send scan request */
      if (m_sock_send(sock, scanrequest, Ustrlen(scanrequest)+1, &errstr) < 0)
        return m_errlog_defer(scanent, CUS callout_address, errstr);

      /* wait for result */
      if (!recv_len(sock, tmpbuf, 2, tmo))
        return m_errlog_defer_3(scanent, CUS callout_address,
                            US"unable to read 2 bytes from socket.", sock);

      /* get errorcode from one nibble */
      kav_rc = tmpbuf[ test_byte_order()==LITTLE_MY_ENDIAN ? 0 : 1 ] & 0x0F;
      switch(kav_rc)
      {
      case 5: case 6: /* improper kavdaemon configuration */
        return m_errlog_defer_3(scanent, CUS callout_address,
                US"please reconfigure kavdaemon to NOT disinfect or remove infected files.",
                sock);
      case 1:
        return m_errlog_defer_3(scanent, CUS callout_address,
                US"reported 'scanning not completed' (code 1).", sock);
      case 7:
        return m_errlog_defer_3(scanent, CUS callout_address,
                US"reported 'kavdaemon damaged' (code 7).", sock);
      }

      /* code 8 is not handled, since it is ambiguous. It appears mostly on
      bounces where part of a file has been cut off */

      /* "virus found" return codes (2-4) */
      if (kav_rc > 1 && kav_rc < 5)
        {
        int report_flag = 0;

        /* setup default virus name */
        malware_name = US"unknown";

        report_flag = tmpbuf[ test_byte_order() == LITTLE_MY_ENDIAN ? 1 : 0 ];

        /* read the report, if available */
        if (report_flag == 1)
          {
          /* read report size */
          if (!recv_len(sock, &kav_reportlen, 4, tmo))
            return m_errlog_defer_3(scanent, CUS callout_address,
                  US"cannot read report size", sock);

          /* it's possible that avp returns av_buffer[1] == 1 but the
          reportsize is 0 (!?) */
          if (kav_reportlen > 0)
            {
            /* set up match regex, depends on retcode */
            if (kav_rc == 3)
              {
              if (!kav_re_sus) kav_re_sus = m_pcre_compile(kav_re_sus_str, &errstr);
              kav_re = kav_re_sus;
              }
            else
              {
              if (!kav_re_inf) kav_re_inf = m_pcre_compile(kav_re_inf_str, &errstr);
              kav_re = kav_re_inf;
              }

            /* read report, linewise.  Using size from stream to read amount of data
            from same stream is safe enough. */
            /* coverity[tainted_data] */
            while (kav_reportlen > 0)
              {
              if ((bread = recv_line(sock, tmpbuf, sizeof(tmpbuf), tmo)) < 0)
                break;
              kav_reportlen -= bread+1;

              /* try matcher on the line, grab substring */
              if ((malware_name = m_pcre_exec(kav_re, tmpbuf)))
                break;
              }
            }
          }
        }
      else /* no virus found */
        malware_name = NULL;

      break;
      }

    case M_CMDL: /* "cmdline" scanner type ---------------------------------- */
      {
      const uschar *cmdline_scanner = scanner_options;
      const pcre *cmdline_trigger_re;
      const pcre *cmdline_regex_re;
      uschar * file_name;
      uschar * commandline;
      void (*eximsigchld)(int);
      void (*eximsigpipe)(int);
      FILE *scanner_out = NULL;
      int scanner_fd;
      FILE *scanner_record = NULL;
      uschar linebuffer[32767];
      int rcnt;
      int trigger = 0;
      uschar *p;

      if (!cmdline_scanner)
        return m_errlog_defer(scanent, NULL, errstr);

      /* find scanner output trigger */
      cmdline_trigger_re = m_pcre_nextinlist(&av_scanner_work, &sep,
                                "missing trigger specification", &errstr);
      if (!cmdline_trigger_re)
        return m_errlog_defer(scanent, NULL, errstr);

      /* find scanner name regex */
      cmdline_regex_re = m_pcre_nextinlist(&av_scanner_work, &sep,
                          "missing virus name regex specification", &errstr);
      if (!cmdline_regex_re)
        return m_errlog_defer(scanent, NULL, errstr);

      /* prepare scanner call; despite the naming, file_name holds a directory
      name which is documented as the value given to %s. */

      file_name = string_copy(eml_filename);
      p = Ustrrchr(file_name, '/');
      if (p)
        *p = '\0';
      commandline = string_sprintf(CS cmdline_scanner, file_name);

      /* redirect STDERR too */
      commandline = string_sprintf("%s 2>&1", commandline);

      DEBUG(D_acl) debug_printf_indent("Malware scan: issuing %s scan [%s]\n",
              scanner_name, commandline);

      /* store exims signal handlers */
      eximsigchld = signal(SIGCHLD,SIG_DFL);
      eximsigpipe = signal(SIGPIPE,SIG_DFL);

      if (!(scanner_out = popen(CS commandline,"r")))
        {
        int err = errno;
        signal(SIGCHLD,eximsigchld); signal(SIGPIPE,eximsigpipe);
        return m_errlog_defer(scanent, NULL,
          string_sprintf("call (%s) failed: %s.", commandline, strerror(err)));
        }
      scanner_fd = fileno(scanner_out);

      file_name = string_sprintf("%s/%s_scanner_output", eml_dir, message_id);

      if (!(scanner_record = modefopen(file_name, "wb", SPOOL_MODE)))
        {
        int err = errno;
        (void) pclose(scanner_out);
        signal(SIGCHLD,eximsigchld); signal(SIGPIPE,eximsigpipe);
        return m_errlog_defer(scanent, NULL, string_sprintf(
            "opening scanner output file (%s) failed: %s.",
            file_name, strerror(err)));
        }

      /* look for trigger while recording output */
      while ((rcnt = recv_line(scanner_fd, linebuffer,
                      sizeof(linebuffer), tmo)))
        {
        if (rcnt < 0)
          {
          int err = errno;
          if (rcnt == -1)
            break;
          (void) pclose(scanner_out);
          signal(SIGCHLD,eximsigchld); signal(SIGPIPE,eximsigpipe);
          return m_errlog_defer(scanent, NULL, string_sprintf(
              "unable to read from scanner (%s): %s",
              commandline, strerror(err)));
          }

        if (Ustrlen(linebuffer) > fwrite(linebuffer, 1, Ustrlen(linebuffer), scanner_record))
          {
          /* short write */
          (void) pclose(scanner_out);
          signal(SIGCHLD,eximsigchld); signal(SIGPIPE,eximsigpipe);
          return m_errlog_defer(scanent, NULL, string_sprintf(
            "short write on scanner output file (%s).", file_name));
          }
        putc('\n', scanner_record);
        /* try trigger match */
        if (  !trigger
           && regex_match_and_setup(cmdline_trigger_re, linebuffer, 0, -1)
           )
          trigger = 1;
        }

      (void)fclose(scanner_record);
      sep = pclose(scanner_out);
      signal(SIGCHLD,eximsigchld); signal(SIGPIPE,eximsigpipe);
      if (sep != 0)
          return m_errlog_defer(scanent, NULL, 
              sep == -1
              ? string_sprintf("running scanner failed: %s", strerror(sep))
              : string_sprintf("scanner returned error code: %d", sep));

      if (trigger)
        {
        uschar * s;
        /* setup default virus name */
        malware_name = US"unknown";

        /* re-open the scanner output file, look for name match */
        scanner_record = fopen(CS file_name, "rb");
        while (fgets(CS linebuffer, sizeof(linebuffer), scanner_record))
          {
          /* try match */
          if ((s = m_pcre_exec(cmdline_regex_re, linebuffer)))
            malware_name = s;
          }
        (void)fclose(scanner_record);
        }
      else /* no virus found */
        malware_name = NULL;
      break;
      } /* cmdline */

    case M_SOPHIE: /* "sophie" scanner type --------------------------------- */
      {
      int bread = 0;
      uschar *p;
      uschar * file_name;
      uschar av_buffer[1024];

      /* pass the scan directory to sophie */
      file_name = string_copy(eml_filename);
      if ((p = Ustrrchr(file_name, '/')))
        *p = '\0';

      DEBUG(D_acl) debug_printf_indent("Malware scan: issuing %s scan [%s]\n",
          scanner_name, scanner_options);

      if (  write(sock, file_name, Ustrlen(file_name)) < 0
         || write(sock, "\n", 1) != 1
         )
        return m_errlog_defer_3(scanent, CUS callout_address,
          string_sprintf("unable to write to UNIX socket (%s)", scanner_options),
          sock);

      /* wait for result */
      memset(av_buffer, 0, sizeof(av_buffer));
      if ((bread = ip_recv(sock, av_buffer, sizeof(av_buffer), tmo-time(NULL))) <= 0)
        return m_errlog_defer_3(scanent, CUS callout_address,
          string_sprintf("unable to read from UNIX socket (%s)", scanner_options),
          sock);

      /* infected ? */
      if (av_buffer[0] == '1') {
        uschar * s = Ustrchr(av_buffer, '\n');
        if (s)
          *s = '\0';
        malware_name = string_copy(&av_buffer[2]);
      }
      else if (!strncmp(CS av_buffer, "-1", 2))
        return m_errlog_defer_3(scanent, CUS callout_address,
                US"scanner reported error", sock);
      else /* all ok, no virus */
        malware_name = NULL;

      break;
      }

    case M_CLAMD: /* "clamd" scanner type ----------------------------------- */
      {
/* This code was originally contributed by David Saez */
/* There are three scanning methods available to us:
*  (1) Use the SCAN command, pointing to a file in the filesystem
*  (2) Use the STREAM command, send the data on a separate port
*  (3) Use the zINSTREAM command, send the data inline
* The zINSTREAM command was introduced with ClamAV 0.95, which marked
* STREAM deprecated; see: http://wiki.clamav.net/bin/view/Main/UpgradeNotes095
* In Exim, we use SCAN if using a Unix-domain socket or explicitly told that
* the TCP-connected daemon is actually local; otherwise we use zINSTREAM unless
* WITH_OLD_CLAMAV_STREAM is defined.
* See Exim bug 926 for details.  */

      uschar *p, *vname, *result_tag;
      int bread=0;
      uschar * file_name;
      uschar av_buffer[1024];
      uschar *hostname = US"";
      host_item connhost;
      uschar *clamav_fbuf;
      int clam_fd, result;
      off_t fsize;
      unsigned int fsize_uint;
      BOOL use_scan_command = FALSE;
      clamd_address * cv[MAX_CLAMD_SERVERS];
      int num_servers = 0;
#ifdef WITH_OLD_CLAMAV_STREAM
      unsigned int port;
      uschar av_buffer2[1024];
      int sockData;
#else
      uint32_t send_size, send_final_zeroblock;
#endif

      /*XXX if unixdomain socket, only one server supported. Needs fixing;
      there's no reason we should not mix local and remote servers */

      if (*scanner_options == '/')
        {
        clamd_address * cd;
        const uschar * sublist;
        int subsep = ' ';

        /* Local file; so we def want to use_scan_command and don't want to try
         * passing IP/port combinations */
        use_scan_command = TRUE;
        cd = (clamd_address *) store_get(sizeof(clamd_address));

        /* extract socket-path part */
        sublist = scanner_options;
        cd->hostspec = string_nextinlist(&sublist, &subsep, NULL, 0);

        /* parse options */
        if (clamd_option(cd, sublist, &subsep) != OK)
          return m_errlog_defer(scanent, NULL,
            string_sprintf("bad option '%s'", scanner_options));
        cv[0] = cd;
        }
      else
        {
        /* Go through the rest of the list of host/port and construct an array
         * of servers to try. The first one is the bit we just passed from
         * scanner_options so process that first and then scan the remainder of
         * the address buffer */
        do
          {
          clamd_address * cd;
          const uschar * sublist;
          int subsep = ' ';
          uschar * s;

          /* The 'local' option means use the SCAN command over the network
           * socket (ie common file storage in use) */
          /*XXX we could accept this also as a local option? */
          if (strcmpic(scanner_options, US"local") == 0)
            {
            use_scan_command = TRUE;
            continue;
            }

          cd = (clamd_address *) store_get(sizeof(clamd_address));

          /* extract host and port part */
          sublist = scanner_options;
          if (!(cd->hostspec = string_nextinlist(&sublist, &subsep, NULL, 0)))
            {
            (void) m_errlog_defer(scanent, NULL, 
                      string_sprintf("missing address: '%s'", scanner_options));
            continue;
            }
          if (!(s = string_nextinlist(&sublist, &subsep, NULL, 0)))
            {
            (void) m_errlog_defer(scanent, NULL, 
                      string_sprintf("missing port: '%s'", scanner_options));
            continue;
            }
          cd->tcp_port = atoi(CS s);

          /* parse options */
          /*XXX should these options be common over scanner types? */
          if (clamd_option(cd, sublist, &subsep) != OK)
            return m_errlog_defer(scanent, NULL,
              string_sprintf("bad option '%s'", scanner_options));

          cv[num_servers++] = cd;
          if (num_servers >= MAX_CLAMD_SERVERS)
            {
            (void) m_errlog_defer(scanent, NULL,
                  US"More than " MAX_CLAMD_SERVERS_S " clamd servers "
                  "specified; only using the first " MAX_CLAMD_SERVERS_S );
            break;
            }
          } while ((scanner_options = string_nextinlist(&av_scanner_work, &sep,
                                        NULL, 0)));

        /* check if we have at least one server */
        if (!num_servers)
          return m_errlog_defer(scanent, NULL,
            US"no useable server addresses in malware configuration option.");
        }

      /* See the discussion of response formats below to see why we really
      don't like colons in filenames when passing filenames to ClamAV. */
      if (use_scan_command && Ustrchr(eml_filename, ':'))
        return m_errlog_defer(scanent, NULL,
          string_sprintf("local/SCAN mode incompatible with" \
            " : in path to email filename [%s]", eml_filename));

      /* We have some network servers specified */
      if (num_servers)
        {
        /* Confirmed in ClamAV source (0.95.3) that the TCPAddr option of clamd
         * only supports AF_INET, but we should probably be looking to the
         * future and rewriting this to be protocol-independent anyway. */

        while (num_servers > 0)
          {
          int i = random_number( num_servers );
          clamd_address * cd = cv[i];

          DEBUG(D_acl) debug_printf_indent("trying server name %s, port %u\n",
                         cd->hostspec, cd->tcp_port);

          /* Lookup the host. This is to ensure that we connect to the same IP
           * on both connections (as one host could resolve to multiple ips) */
          for (;;)
            {
            sock= m_tcpsocket(cd->hostspec, cd->tcp_port, &connhost, &errstr);
            if (sock >= 0)
              {
              /* Connection successfully established with a server */
              hostname = cd->hostspec;
              break;
              }
            if (cd->retry <= 0) break;
            while (cd->retry > 0) cd->retry = sleep(cd->retry);
            }
          if (sock >= 0)
            break;

          (void) m_errlog_defer(scanent, CUS callout_address, errstr);

          /* Remove the server from the list. XXX We should free the memory */
          num_servers--;
          for (; i < num_servers; i++)
            cv[i] = cv[i+1];
          }

        if (num_servers == 0)
          return m_errlog_defer(scanent, NULL, US"all servers failed");
        }
      else
        for (;;)
          {
          if ((sock = ip_unixsocket(cv[0]->hostspec, &errstr)) >= 0)
            {
            hostname = cv[0]->hostspec;
            break;
            }
          if (cv[0]->retry <= 0)
            return m_errlog_defer(scanent, CUS callout_address, errstr);
          while (cv[0]->retry > 0) cv[0]->retry = sleep(cv[0]->retry);
          }

      /* have socket in variable "sock"; command to use is semi-independent of
       * the socket protocol.  We use SCAN if is local (either Unix/local
       * domain socket, or explicitly told local) else we stream the data.
       * How we stream the data depends upon how we were built.  */

      if (!use_scan_command)
        {
#ifdef WITH_OLD_CLAMAV_STREAM
        /* "STREAM\n" command, get back a "PORT <N>\n" response, send data to
         * that port on a second connection; then in the scan-method-neutral
         * part, read the response back on the original connection. */

        DEBUG(D_acl) debug_printf_indent(
            "Malware scan: issuing %s old-style remote scan (PORT)\n",
            scanner_name);

        /* Pass the string to ClamAV (7 = "STREAM\n") */
        if (m_sock_send(sock, US"STREAM\n", 7, &errstr) < 0)
          return m_errlog_defer(scanent, CUS callout_address, errstr);

        memset(av_buffer2, 0, sizeof(av_buffer2));
        bread = ip_recv(sock, av_buffer2, sizeof(av_buffer2), tmo-time(NULL));

        if (bread < 0)
          return m_errlog_defer_3(scanent, CUS callout_address,
            string_sprintf("unable to read PORT from socket (%s)",
                strerror(errno)),
            sock);

        if (bread == sizeof(av_buffer2))
          return m_errlog_defer_3(scanent, CUS callout_address,
                  "buffer too small", sock);

        if (!(*av_buffer2))
          return m_errlog_defer_3(scanent, CUS callout_address,
                  "ClamAV returned null", sock);

        av_buffer2[bread] = '\0';
        if( sscanf(CS av_buffer2, "PORT %u\n", &port) != 1 )
          return m_errlog_defer_3(scanent, CUS callout_address,
            string_sprintf("Expected port information from clamd, got '%s'",
              av_buffer2),
            sock);

        sockData = m_tcpsocket(connhost.address, port, NULL, &errstr);
        if (sockData < 0)
          return m_errlog_defer_3(scanent, CUS callout_address, errstr, sock);

# define CLOSE_SOCKDATA (void)close(sockData)
#else /* WITH_OLD_CLAMAV_STREAM not defined */
        /* New protocol: "zINSTREAM\n" followed by a sequence of <length><data>
        chunks, <n> a 4-byte number (network order), terminated by a zero-length
        chunk. */

        DEBUG(D_acl) debug_printf_indent(
            "Malware scan: issuing %s new-style remote scan (zINSTREAM)\n",
            scanner_name);

        /* Pass the string to ClamAV (10 = "zINSTREAM\0") */
        if (send(sock, "zINSTREAM", 10, 0) < 0)
          return m_errlog_defer_3(scanent, CUS hostname,
            string_sprintf("unable to send zINSTREAM to socket (%s)",
              strerror(errno)),
            sock);

# define CLOSE_SOCKDATA /**/
#endif

        /* calc file size */
        if ((clam_fd = open(CS eml_filename, O_RDONLY)) < 0)
          {
          int err = errno;
          CLOSE_SOCKDATA;
          return m_errlog_defer_3(scanent, NULL,
            string_sprintf("can't open spool file %s: %s",
              eml_filename, strerror(err)),
            sock);
          }
        if ((fsize = lseek(clam_fd, 0, SEEK_END)) < 0)
          {
          int err = errno;
          CLOSE_SOCKDATA; (void)close(clam_fd);
          return m_errlog_defer_3(scanent, NULL,
            string_sprintf("can't seek spool file %s: %s",
              eml_filename, strerror(err)),
            sock);
          }
        fsize_uint = (unsigned int) fsize;
        if ((off_t)fsize_uint != fsize)
          {
          CLOSE_SOCKDATA; (void)close(clam_fd);
          return m_errlog_defer_3(scanent, NULL,
            string_sprintf("seeking spool file %s, size overflow",
              eml_filename),
            sock);
          }
        lseek(clam_fd, 0, SEEK_SET);

        if (!(clamav_fbuf = US malloc(fsize_uint)))
          {
          CLOSE_SOCKDATA; (void)close(clam_fd);
          return m_errlog_defer_3(scanent, NULL,
            string_sprintf("unable to allocate memory %u for file (%s)",
              fsize_uint, eml_filename),
            sock);
          }

        if ((result = read(clam_fd, clamav_fbuf, fsize_uint)) < 0)
          {
          int err = errno;
          free(clamav_fbuf); CLOSE_SOCKDATA; (void)close(clam_fd);
          return m_errlog_defer_3(scanent, NULL,
            string_sprintf("can't read spool file %s: %s",
              eml_filename, strerror(err)),
            sock);
          }
        (void)close(clam_fd);

        /* send file body to socket */
#ifdef WITH_OLD_CLAMAV_STREAM
        if (send(sockData, clamav_fbuf, fsize_uint, 0) < 0)
          {
          free(clamav_fbuf); CLOSE_SOCKDATA;
          return m_errlog_defer_3(scanent, NULL,
            string_sprintf("unable to send file body to socket (%s:%u)",
              hostname, port),
            sock);
          }
#else
        send_size = htonl(fsize_uint);
        send_final_zeroblock = 0;
        if ((send(sock, &send_size, sizeof(send_size), 0) < 0) ||
            (send(sock, clamav_fbuf, fsize_uint, 0) < 0) ||
            (send(sock, &send_final_zeroblock, sizeof(send_final_zeroblock), 0) < 0))
          {
          free(clamav_fbuf);
          return m_errlog_defer_3(scanent, NULL,
            string_sprintf("unable to send file body to socket (%s)", hostname),
            sock);
          }
#endif

        free(clamav_fbuf);

        CLOSE_SOCKDATA;
#undef CLOSE_SOCKDATA
        }
      else
        { /* use scan command */
        /* Send a SCAN command pointing to a filename; then in the then in the
         * scan-method-neutral part, read the response back */

/* ================================================================= */

        /* Prior to the reworking post-Exim-4.72, this scanned a directory,
        which dates to when ClamAV needed us to break apart the email into the
        MIME parts (eg, with the now deprecated demime condition coming first).
        Some time back, ClamAV gained the ability to deconstruct the emails, so
        doing this would actually have resulted in the mail attachments being
        scanned twice, in the broken out files and from the original .eml.
        Since ClamAV now handles emails (and has for quite some time) we can
        just use the email file itself. */
        /* Pass the string to ClamAV (7 = "SCAN \n" + \0) */
        file_name = string_sprintf("SCAN %s\n", eml_filename);

        DEBUG(D_acl) debug_printf_indent(
            "Malware scan: issuing %s local-path scan [%s]\n",
            scanner_name, scanner_options);

        if (send(sock, file_name, Ustrlen(file_name), 0) < 0)
          return m_errlog_defer_3(scanent, CUS callout_address,
            string_sprintf("unable to write to socket (%s)", strerror(errno)),
            sock);

        /* Do not shut down the socket for writing; a user report noted that
         * clamd 0.70 does not react well to this. */
        }
      /* Commands have been sent, no matter which scan method or connection
       * type we're using; now just read the result, independent of method. */

      /* Read the result */
      memset(av_buffer, 0, sizeof(av_buffer));
      bread = ip_recv(sock, av_buffer, sizeof(av_buffer), tmo-time(NULL));
      (void)close(sock);
      sock = -1;

      if (bread <= 0)
        return m_errlog_defer(scanent, CUS callout_address,
          string_sprintf("unable to read from socket (%s)",
          errno == 0 ? "EOF" : strerror(errno)));

      if (bread == sizeof(av_buffer))
        return m_errlog_defer(scanent, CUS callout_address,
                US"buffer too small");
      /* We're now assured of a NULL at the end of av_buffer */

      /* Check the result. ClamAV returns one of two result formats.
      In the basic mode, the response is of the form:
        infected: -> "<filename>: <virusname> FOUND"
        not-infected: -> "<filename>: OK"
        error: -> "<filename>: <errcode> ERROR
      If the ExtendedDetectionInfo option has been turned on, then we get:
        "<filename>: <virusname>(<virushash>:<virussize>) FOUND"
      for the infected case.  Compare:
/tmp/eicar.com: Eicar-Test-Signature FOUND
/tmp/eicar.com: Eicar-Test-Signature(44d88612fea8a8f36de82e1278abb02f:68) FOUND

      In the streaming case, clamd uses the filename "stream" which you should
      be able to verify with { ktrace clamdscan --stream /tmp/eicar.com }.  (The
      client app will replace "stream" with the original filename before returning
      results to stdout, but the trace shows the data).

      We will assume that the pathname passed to clamd from Exim does not contain
      a colon.  We will have whined loudly above if the eml_filename does (and we're
      passing a filename to clamd). */

      if (!(*av_buffer))
        return m_errlog_defer(scanent, CUS callout_address,
                US"ClamAV returned null");

      /* strip newline at the end (won't be present for zINSTREAM)
      (also any trailing whitespace, which shouldn't exist, but we depend upon
      this below, so double-check) */
      p = av_buffer + Ustrlen(av_buffer) - 1;
      if (*p == '\n') *p = '\0';

      DEBUG(D_acl) debug_printf_indent("Malware response: %s\n", av_buffer);

      while (isspace(*--p) && (p > av_buffer))
        *p = '\0';
      if (*p) ++p;

      /* colon in returned output? */
      if(!(p = Ustrchr(av_buffer,':')))
        return m_errlog_defer(scanent, CUS callout_address, string_sprintf(
                  "ClamAV returned malformed result (missing colon): %s",
                  av_buffer));

      /* strip filename */
      while (*p && isspace(*++p)) /**/;
      vname = p;

      /* It would be bad to encounter a virus with "FOUND" in part of the name,
      but we should at least be resistant to it. */
      p = Ustrrchr(vname, ' ');
      result_tag = p ? p+1 : vname;

      if (Ustrcmp(result_tag, "FOUND") == 0)
        {
        /* p should still be the whitespace before the result_tag */
        while (isspace(*p)) --p;
        *++p = '\0';
        /* Strip off the extended information too, which will be in parens
        after the virus name, with no intervening whitespace. */
        if (*--p == ')')
          {
          /* "(hash:size)", so previous '(' will do; if not found, we have
          a curious virus name, but not an error. */
          p = Ustrrchr(vname, '(');
          if (p)
            *p = '\0';
          }
        malware_name = string_copy(vname);
        DEBUG(D_acl) debug_printf_indent("Malware found, name \"%s\"\n", malware_name);

        }
      else if (Ustrcmp(result_tag, "ERROR") == 0)
        return m_errlog_defer(scanent, CUS callout_address,
          string_sprintf("ClamAV returned: %s", av_buffer));

      else if (Ustrcmp(result_tag, "OK") == 0)
        {
        /* Everything should be OK */
        malware_name = NULL;
        DEBUG(D_acl) debug_printf_indent("Malware not found\n");

        }
      else
        return m_errlog_defer(scanent, CUS callout_address,
          string_sprintf("unparseable response from ClamAV: {%s}", av_buffer));

      break;
      } /* clamd */

    case M_SOCK: /* "sock" scanner type ------------------------------------- */
    /* This code was derived by Martin Poole from the clamd code contributed
       by David Saez and the cmdline code
    */
      {
      int bread;
      uschar * commandline;
      uschar av_buffer[1024];
      uschar * linebuffer;
      uschar * sockline_scanner;
      uschar sockline_scanner_default[] = "%s\n";
      const pcre *sockline_trig_re;
      const pcre *sockline_name_re;

      /* find scanner command line */
      if ((sockline_scanner = string_nextinlist(&av_scanner_work, &sep,
                                          NULL, 0)))
      { /* check for no expansions apart from one %s */
        uschar * s = Ustrchr(sockline_scanner, '%');
        if (s++)
          if ((*s != 's' && *s != '%') || Ustrchr(s+1, '%'))
            return m_errlog_defer_3(scanent, NULL,
                                  US"unsafe sock scanner call spec", sock);
      }
      else
        sockline_scanner = sockline_scanner_default;

      /* find scanner output trigger */
      sockline_trig_re = m_pcre_nextinlist(&av_scanner_work, &sep,
                                "missing trigger specification", &errstr);
      if (!sockline_trig_re)
        return m_errlog_defer_3(scanent, NULL, errstr, sock);

      /* find virus name regex */
      sockline_name_re = m_pcre_nextinlist(&av_scanner_work, &sep,
                          "missing virus name regex specification", &errstr);
      if (!sockline_name_re)
        return m_errlog_defer_3(scanent, NULL, errstr, sock);

      /* prepare scanner call - security depends on expansions check above */
      commandline = string_sprintf( CS sockline_scanner, CS eml_filename);


      /* Pass the command string to the socket */
      if (m_sock_send(sock, commandline, Ustrlen(commandline), &errstr) < 0)
        return m_errlog_defer(scanent, CUS callout_address, errstr);

      /* Read the result */
      bread = ip_recv(sock, av_buffer, sizeof(av_buffer), tmo-time(NULL));

      if (bread <= 0)
        return m_errlog_defer_3(scanent, CUS callout_address,
          string_sprintf("unable to read from socket (%s)", strerror(errno)),
          sock);

      if (bread == sizeof(av_buffer))
        return m_errlog_defer_3(scanent, CUS callout_address,
                US"buffer too small", sock);
      av_buffer[bread] = '\0';
      linebuffer = string_copy(av_buffer);

      /* try trigger match */
      if (regex_match_and_setup(sockline_trig_re, linebuffer, 0, -1))
        {
        if (!(malware_name = m_pcre_exec(sockline_name_re, av_buffer)))
          malware_name = US "unknown";
        }
      else /* no virus found */
        malware_name = NULL;
      break;
      }

    case M_MKSD: /* "mksd" scanner type ------------------------------------- */
      {
      char *mksd_options_end;
      int mksd_maxproc = 1;  /* default, if no option supplied */
      int retval;

      if (scanner_options)
        {
        mksd_maxproc = (int)strtol(CS scanner_options, &mksd_options_end, 10);
        if (  *scanner_options == '\0'
           || *mksd_options_end != '\0'
           || mksd_maxproc < 1
           || mksd_maxproc > 32
           )
          return m_errlog_defer(scanent, CUS callout_address,
            string_sprintf("invalid option '%s'", scanner_options));
        }

      if((sock = ip_unixsocket(US "/var/run/mksd/socket", &errstr)) < 0)
        return m_errlog_defer(scanent, CUS callout_address, errstr);

      malware_name = NULL;

      DEBUG(D_acl) debug_printf_indent("Malware scan: issuing %s scan\n", scanner_name);

      if ((retval = mksd_scan_packed(scanent, sock, eml_filename, tmo)) != OK)
        {
        close (sock);
        return retval;
        }
      break;
      }

    case M_AVAST: /* "avast" scanner type ----------------------------------- */
      {
      int ovector[1*3];
      uschar buf[1024];
      uschar * scanrequest;
      enum {AVA_HELO, AVA_OPT, AVA_RSP, AVA_DONE} avast_stage;
      int nread;

      /* According to Martin Tuma @avast the protocol uses "escaped
      whitespace", that is, every embedded whitespace is backslash
      escaped, as well as backslash is protected by backslash.
      The returned lines contain the name of the scanned file, a tab
      and the [ ] marker.
      [+] - not infected
      [L] - infected
      [E] - some error occured
      Such marker follows the first non-escaped TAB.  */
      if (  (  !ava_re_clean
            && !(ava_re_clean = m_pcre_compile(ava_re_clean_str, &errstr)))
         || (  !ava_re_virus
            && !(ava_re_virus = m_pcre_compile(ava_re_virus_str, &errstr)))
         )
        return malware_errlog_defer(errstr);

      /* wait for result */
      for (avast_stage = AVA_HELO;
           (nread = recv_line(sock, buf, sizeof(buf), tmo)) > 0;
          )
        {
        int slen = Ustrlen(buf);
        if (slen >= 1)
          {
          DEBUG(D_acl) debug_printf_indent("got from avast: %s\n", buf);
          switch (avast_stage)
            {
            case AVA_HELO:
              if (Ustrncmp(buf, "220", 3) != 0)
                goto endloop;                   /* require a 220 */
              goto sendreq;

            case AVA_OPT:
              if (Ustrncmp(buf, "210", 3) == 0)
                break;                          /* ignore 210 responses */
              if (Ustrncmp(buf, "200", 3) != 0)
                goto endloop;                   /* require a 200 */

            sendreq:
              {
              int len;
              /* Check for another option to send. Newline-terminate it. */
              if ((scanrequest = string_nextinlist(&av_scanner_work, &sep,
                                NULL, 0)))
                {
                scanrequest = string_sprintf("%s\n", scanrequest);
                avast_stage = AVA_OPT;          /* just sent option */
                }
              else
                {
                scanrequest = string_sprintf("SCAN %s\n", eml_dir);
                avast_stage = AVA_RSP;          /* just sent command */
                }

              /* send config-cmd or scan-request to socket */
              len = Ustrlen(scanrequest);
              if (send(sock, scanrequest, len, 0) < 0)
                {
                scanrequest[len-1] = '\0';
                return m_errlog_defer_3(scanent, CUS callout_address, string_sprintf(
                      "unable to send request '%s' to socket (%s): %s",
                      scanrequest, scanner_options, strerror(errno)), sock);
                }
              break;
              }

            case AVA_RSP:
              if (Ustrncmp(buf, "210", 3) == 0)
                break;  /* ignore the "210 SCAN DATA" message */

              if (pcre_exec(ava_re_clean, NULL, CS buf, slen,
                    0, 0, ovector, nelements(ovector)) > 0)
                break;

              if ((malware_name = m_pcre_exec(ava_re_virus, buf)))
                { /* remove backslash in front of [whitespace|backslash] */
                uschar * p, * p0;
                for (p = malware_name; *p; ++p)
                  if (*p == '\\' && (isspace(p[1]) || p[1] == '\\'))
                    for (p0 = p; *p0; ++p0) *p0 = p0[1];

                avast_stage = AVA_DONE;
                goto endloop;
                }

              if (Ustrncmp(buf, "200 SCAN OK", 11) == 0)
                { /* we're done finally */
                if (send(sock, "QUIT\n", 5, 0) < 0) /* courtesy */
                  return m_errlog_defer_3(scanent, CUS callout_address,
                          string_sprintf(
                              "unable to send quit request to socket (%s): %s",
                              scanner_options, strerror(errno)),
                              sock);
                malware_name = NULL;
                avast_stage = AVA_DONE;
                goto endloop;
                }

              /* here for any unexpected response from the scanner */
              goto endloop;

            case AVA_DONE:      log_write(0, LOG_PANIC, "%s:%d:%s: should not happen",
                            __FILE__, __LINE__, __FUNCTION__);
            }
          }
        }
      endloop:

      switch(avast_stage)
        {
        case AVA_HELO:
        case AVA_OPT:
        case AVA_RSP:   return m_errlog_defer_3(scanent, CUS callout_address,
                            nread >= 0
                            ? string_sprintf(
                                "invalid response from scanner: '%s'", buf)
                            : nread == -1
                            ? US"EOF from scanner"
                            : US"timeout from scanner",
                          sock);
        default:        break;
        }
      break;
      }

    case M_FPROT6D: /* "f-prot6d" scanner type ----------------------------------- */
      {
      int bread;
      uschar * e;
      uschar * linebuffer;
      uschar * scanrequest;
      uschar av_buffer[1024];

      if ((!fprot6d_re_virus && !(fprot6d_re_virus = m_pcre_compile(fprot6d_re_virus_str, &errstr)))
        || (!fprot6d_re_error && !(fprot6d_re_error = m_pcre_compile(fprot6d_re_error_str, &errstr))))
        return malware_errlog_defer(errstr);

      scanrequest = string_sprintf("SCAN FILE %s\n", eml_filename);
      DEBUG(D_acl) debug_printf_indent("Malware scan: issuing %s: %s\n",
        scanner_name, scanrequest);

      if (m_sock_send(sock, scanrequest, Ustrlen(scanrequest), &errstr) < 0)
        return m_errlog_defer(scanent, CUS callout_address, errstr);

      bread = ip_recv(sock, av_buffer, sizeof(av_buffer), tmo-time(NULL));

      if (bread <= 0)
        return m_errlog_defer_3(scanent, CUS callout_address,
          string_sprintf("unable to read from socket (%s)", strerror(errno)),
          sock);

      if (bread == sizeof(av_buffer))
        return m_errlog_defer_3(scanent, CUS callout_address,
          US"buffer too small", sock);

      av_buffer[bread] = '\0';
      linebuffer = string_copy(av_buffer);

      m_sock_send(sock, US"QUIT\n", 5, 0);

      if ((e = m_pcre_exec(fprot6d_re_error, linebuffer)))
        return m_errlog_defer_3(scanent, CUS callout_address,
          string_sprintf("scanner reported error (%s)", e), sock);

      if (!(malware_name = m_pcre_exec(fprot6d_re_virus, linebuffer)))
        malware_name = NULL;

      break;
      }  /* f-prot6d */
  }     /* scanner type switch */

  if (sock >= 0)
    (void) close (sock);
  malware_ok = TRUE;                    /* set "been here, done that" marker */
  }

/* match virus name against pattern (caseless ------->----------v) */
if (malware_name && regex_match_and_setup(re, malware_name, 0, -1))
  {
  DEBUG(D_acl) debug_printf_indent(
      "Matched regex to malware [%s] [%s]\n", malware_re, malware_name);
  return OK;
  }
else
  return FAIL;
}


/*************************************************
*          Scan an email for malware             *
*************************************************/

/* This is the normal interface for scanning an email, which doesn't need a
filename; it's a wrapper around the malware_file function.

Arguments:
  malware_re  match condition for "malware="
  timeout     if nonzero, timeout in seconds

Returns:      Exim message processing code (OK, FAIL, DEFER, ...)
              where true means malware was found (condition applies)
*/
int
malware(const uschar * malware_re, int timeout)
{
int ret = malware_internal(malware_re, NULL, timeout);

if (ret == DEFER) av_failed = TRUE;
return ret;
}


/*************************************************
*          Scan a file for malware               *
*************************************************/

/* This is a test wrapper for scanning an email, which is not used in
normal processing.  Scan any file, using the Exim scanning interface.
This function tampers with various global variables so is unsafe to use
in any other context.

Arguments:
  eml_filename  a file holding the message to be scanned

Returns:        Exim message processing code (OK, FAIL, DEFER, ...)
                where true means malware was found (condition applies)
*/
int
malware_in_file(uschar *eml_filename)
{
uschar message_id_buf[64];
int ret;

/* spool_mbox() assumes various parameters exist, when creating
the relevant directory and the email within */

(void) string_format(message_id_buf, sizeof(message_id_buf),
    "dummy-%d", vaguely_random_number(INT_MAX));
message_id = message_id_buf;
sender_address = US"malware-sender@example.net";
return_path = US"";
recipients_list = NULL;
receive_add_recipient(US"malware-victim@example.net", -1);
enable_dollar_recipients = TRUE;

ret = malware_internal(US"*", eml_filename, 0);

Ustrncpy(spooled_message_id, message_id, sizeof(spooled_message_id));
spool_mbox_ok = 1;

/* don't set no_mbox_unspool; at present, there's no way for it to become
set, but if that changes, then it should apply to these tests too */

unspool_mbox();

/* silence static analysis tools */
message_id = NULL;

return ret;
}


void
malware_init(void)
{
if (!malware_default_re)
  malware_default_re = regex_must_compile(malware_regex_default, FALSE, TRUE);
if (!drweb_re)
  drweb_re = regex_must_compile(drweb_re_str, FALSE, TRUE);
if (!fsec_re)
  fsec_re = regex_must_compile(fsec_re_str, FALSE, TRUE);
if (!kav_re_sus)
  kav_re_sus = regex_must_compile(kav_re_sus_str, FALSE, TRUE);
if (!kav_re_inf)
  kav_re_inf = regex_must_compile(kav_re_inf_str, FALSE, TRUE);
if (!ava_re_clean)
  ava_re_clean = regex_must_compile(ava_re_clean_str, FALSE, TRUE);
if (!ava_re_virus)
  ava_re_virus = regex_must_compile(ava_re_virus_str, FALSE, TRUE);
if (!fprot6d_re_error)
  fprot6d_re_error = regex_must_compile(fprot6d_re_error_str, FALSE, TRUE);
if (!fprot6d_re_virus)
  fprot6d_re_virus = regex_must_compile(fprot6d_re_virus_str, FALSE, TRUE);
}

#endif /*WITH_CONTENT_SCAN*/
/*
 * vi: aw ai sw=2
 */