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

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

/* Copyright (c) The Exim Maintainers 2020 - 2023 */
/* Copyright (c) University of Cambridge 1995 - 2018 */
/* See the file NOTICE for conditions of use and distribution. */
/* SPDX-License-Identifier: GPL-2.0-or-later */

/* Code for receiving a message and setting up spool files. */

#include "exim.h"
#include <setjmp.h>

#ifdef EXPERIMENTAL_DCC
extern int dcc_ok;
#endif

#ifdef SUPPORT_DMARC
# include "dmarc.h"
#endif

/*************************************************
*                Local static variables          *
*************************************************/

static int     data_fd = -1;
static uschar *spool_name = US"";

enum CH_STATE {LF_SEEN, MID_LINE, CR_SEEN};

#ifdef HAVE_LOCAL_SCAN
jmp_buf local_scan_env;         /* error-handling context for local_scan */
unsigned had_local_scan_crash;
unsigned had_local_scan_timeout;
#endif


/*************************************************
*      Non-SMTP character reading functions      *
*************************************************/

/* These are the default functions that are set up in the variables such as
receive_getc initially. They just call the standard functions, passing stdin as
the file. (When SMTP input is occurring, different functions are used by
changing the pointer variables.) */

uschar stdin_buf[4096];
uschar * stdin_inptr = stdin_buf;
uschar * stdin_inend = stdin_buf;

static BOOL
stdin_refill(void)
{
size_t rc = fread(stdin_buf, 1, sizeof(stdin_buf), stdin);
if (rc <= 0)
  {
  if (had_data_timeout)
    {
    fprintf(stderr, "exim: timed out while reading - message abandoned\n");
    log_write(L_lost_incoming_connection,
              LOG_MAIN, "timed out while reading local message");
    receive_bomb_out(US"data-timeout", NULL);   /* Does not return */
    }
  if (had_data_sigint)
    {
    if (filter_test == FTEST_NONE)
      {
      fprintf(stderr, "\nexim: %s received - message abandoned\n",
        had_data_sigint == SIGTERM ? "SIGTERM" : "SIGINT");
      log_write(0, LOG_MAIN, "%s received while reading local message",
        had_data_sigint == SIGTERM ? "SIGTERM" : "SIGINT");
      }
    receive_bomb_out(US"signal-exit", NULL);    /* Does not return */
    }
  return FALSE;
  }
stdin_inend = stdin_buf + rc;
stdin_inptr = stdin_buf;
return TRUE;
}

int
stdin_getc(unsigned lim)
{
if (stdin_inptr >= stdin_inend)
  if (!stdin_refill())
      return EOF;
return *stdin_inptr++;
}


BOOL
stdin_hasc(void)
{
return stdin_inptr < stdin_inend;
}

int
stdin_ungetc(int c)
{
if (stdin_inptr <= stdin_buf)
  log_write(0, LOG_MAIN|LOG_PANIC_DIE, "buffer underflow in stdin_ungetc");

*--stdin_inptr = c;
return c;
}

int
stdin_feof(void)
{
return stdin_hasc() ? FALSE : feof(stdin);
}

int
stdin_ferror(void)
{
return ferror(stdin);
}




/*************************************************
*     Check that a set sender is allowed         *
*************************************************/

/* This function is called when a local caller sets an explicit sender address.
It checks whether this is permitted, which it is for trusted callers.
Otherwise, it must match the pattern(s) in untrusted_set_sender.

Arguments:  the proposed sender address
Returns:    TRUE for a trusted caller
            TRUE if the address has been set, untrusted_set_sender has been
              set, and the address matches something in the list
            FALSE otherwise
*/

BOOL
receive_check_set_sender(const uschar * newsender)
{
const uschar * qnewsender;
if (f.trusted_caller) return TRUE;
if (!newsender || !untrusted_set_sender) return FALSE;
qnewsender = Ustrchr(newsender, '@')
  ? newsender : string_sprintf("%s@%s", newsender, qualify_domain_sender);
return match_address_list_basic(qnewsender, CUSS &untrusted_set_sender, 0) == OK;
}




/*************************************************
*          Read space info for a partition       *
*************************************************/

/* This function is called by receive_check_fs() below, and also by string
expansion for variables such as $spool_space. The field names for the statvfs
structure are macros, because not all OS have F_FAVAIL and it seems tidier to
have macros for F_BAVAIL and F_FILES as well. Some kinds of file system do not
have inodes, and they return -1 for the number available.

Later: It turns out that some file systems that do not have the concept of
inodes return 0 rather than -1. Such systems should also return 0 for the total
number of inodes, so we require that to be greater than zero before returning
an inode count.

Arguments:
  isspool       TRUE for spool partition, FALSE for log partition
  inodeptr      address of int to receive inode count; -1 if there isn't one

Returns:        available on-root space, in kilobytes
                -1 for log partition if there isn't one

All values are -1 if the STATFS functions are not available.
*/

int_eximarith_t
receive_statvfs(BOOL isspool, int *inodeptr)
{
#ifdef HAVE_STATFS
struct STATVFS statbuf;
struct stat dummy;
uschar *path;
uschar *name;
uschar buffer[1024];

/* The spool directory must always exist. */

if (isspool)
  {
  path = spool_directory;
  name = US"spool";
  }

/* Need to cut down the log file path to the directory, and to ignore any
appearance of "syslog" in it. */

else
  {
  int sep = ':';              /* Not variable - outside scripts use */
  const uschar *p = log_file_path;
  name = US"log";

  /* An empty log_file_path means "use the default". This is the same as an
  empty item in a list. */

  if (*p == 0) p = US":";
  /* should never be a tainted list */
  while ((path = string_nextinlist(&p, &sep, buffer, sizeof(buffer))))
    if (Ustrcmp(path, "syslog") != 0)
      break;

  if (path == NULL)  /* No log files */
    {
    *inodeptr = -1;
    return -1;
    }

  /* An empty string means use the default, which is in the spool directory.
  But don't just use the spool directory, as it is possible that the log
  subdirectory has been symbolically linked elsewhere. */

  if (path[0] == 0)
    {
    sprintf(CS buffer, CS"%s/log", CS spool_directory);
    path = buffer;
    }
  else
    {
    uschar *cp;
    if ((cp = Ustrrchr(path, '/')) != NULL) *cp = 0;
    }
  }

/* We now have the path; do the business */

memset(&statbuf, 0, sizeof(statbuf));

if (STATVFS(CS path, &statbuf) != 0)
  if (stat(CS path, &dummy) == -1 && errno == ENOENT)
    {                           /* Can happen on first run after installation */
    *inodeptr = -1;
    return -1;
    }
  else
    {
    log_write(0, LOG_MAIN|LOG_PANIC, "cannot accept message: failed to stat "
      "%s directory %s: %s", name, path, strerror(errno));
    smtp_closedown(US"spool or log directory problem");
    exim_exit(EXIT_FAILURE);
    }

*inodeptr = (statbuf.F_FILES > 0)? statbuf.F_FAVAIL : -1;

/* Disks are getting huge. Take care with computing the size in kilobytes. */

return (int_eximarith_t)(((double)statbuf.F_BAVAIL * (double)statbuf.F_FRSIZE)/1024.0);

#else
/* Unable to find partition sizes in this environment. */

*inodeptr = -1;
return -1;
#endif
}




/*************************************************
*     Check space on spool and log partitions    *
*************************************************/

/* This function is called before accepting a message; if any thresholds are
set, it checks them. If a message_size is supplied, it checks that there is
enough space for that size plus the threshold - i.e. that the message won't
reduce the space to the threshold. Not all OS have statvfs(); for those that
don't, this function always returns TRUE. For some OS the old function and
struct name statfs is used; that is handled by a macro, defined in exim.h.

Arguments:
  msg_size     the (estimated) size of an incoming message

Returns:       FALSE if there isn't enough space, or if the information cannot
                 be obtained
               TRUE if no check was done or there is enough space
*/

BOOL
receive_check_fs(int msg_size)
{
int_eximarith_t space;
int inodes;

if (check_spool_space > 0 || msg_size > 0 || check_spool_inodes > 0)
  {
  space = receive_statvfs(TRUE, &inodes);

  DEBUG(D_receive)
    debug_printf("spool directory space = " PR_EXIM_ARITH "K inodes = %d "
      "check_space = " PR_EXIM_ARITH "K inodes = %d msg_size = %d\n",
      space, inodes, check_spool_space, check_spool_inodes, msg_size);

  if (  space >= 0 && space + msg_size / 1024 < check_spool_space
     || inodes >= 0 && inodes < check_spool_inodes)
    {
    log_write(0, LOG_MAIN, "spool directory space check failed: space="
      PR_EXIM_ARITH " inodes=%d", space, inodes);
    return FALSE;
    }
  }

if (check_log_space > 0 || check_log_inodes > 0)
  {
  space = receive_statvfs(FALSE, &inodes);

  DEBUG(D_receive)
    debug_printf("log directory space = " PR_EXIM_ARITH "K inodes = %d "
      "check_space = " PR_EXIM_ARITH "K inodes = %d\n",
      space, inodes, check_log_space, check_log_inodes);

  if (  space >= 0 && space < check_log_space
     || inodes >= 0 && inodes < check_log_inodes)
    {
    log_write(0, LOG_MAIN, "log directory space check failed: space=" PR_EXIM_ARITH
      " inodes=%d", space, inodes);
    return FALSE;
    }
  }

return TRUE;
}



/*************************************************
*         Bomb out while reading a message       *
*************************************************/

/* The common case of wanting to bomb out is if a SIGTERM or SIGINT is
received, or if there is a timeout. A rarer case might be if the log files are
screwed up and Exim can't open them to record a message's arrival. Handling
that case is done by setting a flag to cause the log functions to call this
function if there is an ultimate disaster. That is why it is globally
accessible.

Arguments:
  reason     text reason to pass to the not-quit ACL
  msg        default SMTP response to give if in an SMTP session
Returns:     it doesn't
*/

void
receive_bomb_out(uschar *reason, uschar *msg)
{
  static BOOL already_bombing_out;
/* The smtp_notquit_exit() below can call ACLs which can trigger recursive
timeouts, if someone has something slow in their quit ACL.  Since the only
things we should be doing are to close down cleanly ASAP, on the second
pass we also close down stuff that might be opened again, before bypassing
the ACL call and exiting. */

/* If spool_name is set, it contains the name of the data file that is being
written. Unlink it before closing so that it cannot be picked up by a delivery
process. Ensure that any header file is also removed. */

if (spool_name[0] != '\0')
  {
  Uunlink(spool_name);
  spool_name[Ustrlen(spool_name) - 1] = 'H';
  Uunlink(spool_name);
  spool_name[0] = '\0';
  }

/* Now close the file if it is open, either as a fd or a stream. */

if (spool_data_file)
  {
  (void)fclose(spool_data_file);
  spool_data_file = NULL;
  }
else if (data_fd >= 0)
  {
  (void)close(data_fd);
  data_fd = -1;
  }

/* Attempt to close down an SMTP connection tidily. For non-batched SMTP, call
smtp_notquit_exit(), which runs the NOTQUIT ACL, if present, and handles the
SMTP response. */

if (!already_bombing_out)
  {
  already_bombing_out = TRUE;
  if (smtp_input)
    {
    if (smtp_batched_input)
      moan_smtp_batch(NULL, "421 %s - message abandoned", msg);  /* No return */
    smtp_notquit_exit(reason, US"421", US"%s %s - closing connection.",
      smtp_active_hostname, msg);
    }
  }

/* Exit from the program (non-BSMTP cases) */

exim_exit(EXIT_FAILURE);
}


/*************************************************
*              Data read timeout                 *
*************************************************/

/* Handler function for timeouts that occur while reading the data that
comprises a message.

Argument:  the signal number
Returns:   nothing
*/

static void
data_timeout_handler(int sig)
{
had_data_timeout = sig;
}



#ifdef HAVE_LOCAL_SCAN
/*************************************************
*              local_scan() timeout              *
*************************************************/

/* Handler function for timeouts that occur while running a local_scan()
function.  Posix recommends against calling longjmp() from a signal-handler,
but the GCC manual says you can so we will, and trust that it's better than
calling probably non-signal-safe funxtions during logging from within the
handler, even with other compilers.

See also https://cwe.mitre.org/data/definitions/745.html which also lists
it as unsafe.

This is all because we have no control over what might be written for a
local-scan function, so cannot sprinkle had-signal checks after each
call-site.  At least with the default "do-nothing" function we won't
ever get here.

Argument:  the signal number
Returns:   nothing
*/

static void
local_scan_timeout_handler(int sig)
{
had_local_scan_timeout = sig;
siglongjmp(local_scan_env, 1);
}



/*************************************************
*            local_scan() crashed                *
*************************************************/

/* Handler function for signals that occur while running a local_scan()
function.

Argument:  the signal number
Returns:   nothing
*/

static void
local_scan_crash_handler(int sig)
{
had_local_scan_crash = sig;
siglongjmp(local_scan_env, 1);
}

#endif /*HAVE_LOCAL_SCAN*/


/*************************************************
*           SIGTERM or SIGINT received           *
*************************************************/

/* Handler for SIGTERM or SIGINT signals that occur while reading the
data that comprises a message.

Argument:  the signal number
Returns:   nothing
*/

static void
data_sigterm_sigint_handler(int sig)
{
had_data_sigint = sig;
}



/*************************************************
*          Add new recipient to list             *
*************************************************/

/* This function builds a list of recipient addresses in argc/argv
format.

Arguments:
  recipient   the next address to add to recipients_list
  pno         parent number for fixed aliases; -1 otherwise

Returns:      nothing
*/

void
receive_add_recipient(const uschar * recipient, int pno)
{
if (recipients_count >= recipients_list_max)
  {
  recipient_item *oldlist = recipients_list;
  int oldmax = recipients_list_max;

  const int safe_recipients_limit = INT_MAX / 2 / sizeof(recipient_item);
  if (recipients_list_max < 0 || recipients_list_max >= safe_recipients_limit)
    {
    log_write(0, LOG_MAIN|LOG_PANIC_DIE, "Too many recipients: %d", recipients_list_max);
    }

  recipients_list_max = recipients_list_max ? 2*recipients_list_max : 50;
  recipients_list = store_get(recipients_list_max * sizeof(recipient_item), GET_UNTAINTED);
  if (oldlist)
    memcpy(recipients_list, oldlist, oldmax * sizeof(recipient_item));
  }

recipients_list[recipients_count].address = recipient;
recipients_list[recipients_count].pno = pno;
#ifdef EXPERIMENTAL_BRIGHTMAIL
recipients_list[recipients_count].bmi_optin = bmi_current_optin;
/* reset optin string pointer for next recipient */
bmi_current_optin = NULL;
#endif
recipients_list[recipients_count].orcpt = NULL;
recipients_list[recipients_count].dsn_flags = 0;
recipients_list[recipients_count++].errors_to = NULL;
}




/*************************************************
*        Send user response message              *
*************************************************/

/* This function is passed a default response code and a user message. It calls
smtp_message_code() to check and possibly modify the response code, and then
calls smtp_respond() to transmit the response. I put this into a function
just to avoid a lot of repetition.

Arguments:
  code         the response code
  user_msg     the user message

Returns:       nothing
*/

#ifndef DISABLE_PRDR
static void
smtp_user_msg(uschar *code, uschar *user_msg)
{
int len = 3;
smtp_message_code(&code, &len, &user_msg, NULL, TRUE);
smtp_respond(code, len, SR_FINAL, user_msg);
}
#endif





/*************************************************
*        Remove a recipient from the list        *
*************************************************/

/* This function is provided for local_scan() to use.

Argument:
  recipient   address to remove

Returns:      TRUE if it did remove something; FALSE otherwise
*/

BOOL
receive_remove_recipient(const uschar * recipient)
{
DEBUG(D_receive) debug_printf("receive_remove_recipient(\"%s\") called\n",
  recipient);
for (int count = 0; count < recipients_count; count++)
  if (Ustrcmp(recipients_list[count].address, recipient) == 0)
    {
    if ((--recipients_count - count) > 0)
      memmove(recipients_list + count, recipients_list + count + 1,
        (recipients_count - count)*sizeof(recipient_item));
    return TRUE;
    }
return FALSE;
}





/* Pause for a while waiting for input.  If none received in that time,
close the logfile, if we had one open; then if we wait for a long-running
datasource (months, in one use-case) log rotation will not leave us holding
the file copy. */

static void
log_close_chk(void)
{
if (!receive_timeout && !receive_hasc())
  {
  struct timeval t;
  timesince(&t, &received_time);
  if (t.tv_sec > 30*60)
    mainlog_close();
  else
    if (poll_one_fd(0, POLLIN, (30*60 - t.tv_sec) * 1000) == 0)
      mainlog_close();
  }
}

/*************************************************
*     Read data portion of a non-SMTP message    *
*************************************************/

/* This function is called to read the remainder of a message (following the
header) when the input is not from SMTP - we are receiving a local message on
a standard input stream. The message is always terminated by EOF, and is also
terminated by a dot on a line by itself if the flag dot_ends is TRUE. Split the
two cases for maximum efficiency.

Ensure that the body ends with a newline. This will naturally be the case when
the termination is "\n.\n" but may not be otherwise. The RFC defines messages
as "sequences of lines" - this of course strictly applies only to SMTP, but
deliveries into BSD-type mailbox files also require it. Exim used to have a
flag for doing this at delivery time, but as it was always set for all
transports, I decided to simplify things by putting the check here instead.

There is at least one MUA (dtmail) that sends CRLF via this interface, and
other programs are known to do this as well. Exim used to have a option for
dealing with this: in July 2003, after much discussion, the code has been
changed to default to treat any of LF, CRLF, and bare CR as line terminators.

However, for the case when a dot on a line by itself terminates a message, the
only recognized terminating sequences before and after the dot are LF and CRLF.
Otherwise, having read EOL . CR, you don't know whether to read another
character or not.

Internally, in messages stored in Exim's spool files, LF is used as the line
terminator. Under the new regime, bare CRs will no longer appear in these
files.

Arguments:
  fout      a FILE to which to write the message

Returns:    One of the END_xxx values indicating why it stopped reading
*/

static int
read_message_data(FILE *fout)
{
int ch_state;
register int ch;
register int linelength = 0;

/* Handle the case when only EOF terminates the message */

if (!f.dot_ends)
  {
  int last_ch = '\n';

  for ( ;
       log_close_chk(), (ch = (receive_getc)(GETC_BUFFER_UNLIMITED)) != EOF;
       last_ch = ch)
    {
    if (ch == 0) body_zerocount++;
    if (last_ch == '\r' && ch != '\n')
      {
      if (linelength > max_received_linelength)
        max_received_linelength = linelength;
      linelength = 0;
      if (fputc('\n', fout) == EOF) return END_WERROR;
      message_size++;
      body_linecount++;
      }
    if (ch == '\r') continue;

    if (fputc(ch, fout) == EOF) return END_WERROR;
    if (ch == '\n')
      {
      if (linelength > max_received_linelength)
        max_received_linelength = linelength;
      linelength = 0;
      body_linecount++;
      }
    else linelength++;
    if (++message_size > thismessage_size_limit) return END_SIZE;
    }

  if (last_ch != '\n')
    {
    if (linelength > max_received_linelength)
      max_received_linelength = linelength;
    if (fputc('\n', fout) == EOF) return END_WERROR;
    message_size++;
    body_linecount++;
    }

  return END_EOF;
  }

/* Handle the case when a dot on a line on its own, or EOF, terminates. */

ch_state = 1;

while (log_close_chk(), (ch = (receive_getc)(GETC_BUFFER_UNLIMITED)) != EOF)
  {
  if (ch == 0) body_zerocount++;
  switch (ch_state)
    {
    case 0:                         /* Normal state (previous char written) */
    if (ch == '\n')
      {
      body_linecount++;
      if (linelength > max_received_linelength)
        max_received_linelength = linelength;
      linelength = -1;
      ch_state = 1;
      }
    else if (ch == '\r')
      { ch_state = 2; continue; }
    break;

    case 1:                         /* After written "\n" */
    if (ch == '.') { ch_state = 3; continue; }
    if (ch == '\r') { ch_state = 2; continue; }
    if (ch == '\n') { body_linecount++; linelength = -1; }
    else ch_state = 0;
    break;

    case 2:
    body_linecount++;               /* After unwritten "\r" */
    if (linelength > max_received_linelength)
      max_received_linelength = linelength;
    if (ch == '\n')
      {
      ch_state = 1;
      linelength = -1;
      }
    else
      {
      if (message_size++, fputc('\n', fout) == EOF) return END_WERROR;
      if (ch == '\r') continue;
      ch_state = 0;
      linelength = 0;
      }
    break;

    case 3:                         /* After "\n." (\n written, dot not) */
    if (ch == '\n') return END_DOT;
    if (ch == '\r') { ch_state = 4; continue; }
    message_size++;
    linelength++;
    if (fputc('.', fout) == EOF) return END_WERROR;
    ch_state = 0;
    break;

    case 4:                         /* After "\n.\r" (\n written, rest not) */
    if (ch == '\n') return END_DOT;
    message_size += 2;
    body_linecount++;
    if (fputs(".\n", fout) == EOF) return END_WERROR;
    if (ch == '\r') { ch_state = 2; continue; }
    ch_state = 0;
    break;
    }

  linelength++;
  if (fputc(ch, fout) == EOF) return END_WERROR;
  if (++message_size > thismessage_size_limit) return END_SIZE;
  }

/* Get here if EOF read. Unless we have just written "\n", we need to ensure
the message ends with a newline, and we must also write any characters that
were saved up while testing for an ending dot. */

if (ch_state != 1)
  {
  static uschar *ends[] = { US"\n", NULL, US"\n", US".\n", US".\n" };
  if (fputs(CS ends[ch_state], fout) == EOF) return END_WERROR;
  message_size += Ustrlen(ends[ch_state]);
  body_linecount++;
  }

return END_EOF;
}




/*************************************************
*      Read data portion of an SMTP message      *
*************************************************/

/* This function is called to read the remainder of an SMTP message (after the
headers), or to skip over it when an error has occurred. In this case, the
output file is passed as NULL.

If any line begins with a dot, that character is skipped. The input should only
be successfully terminated by CR LF . CR LF unless it is local (non-network)
SMTP, in which case the CRs are optional, but...

FUDGE: It seems that sites on the net send out messages with just LF
terminators, despite the warnings in the RFCs, and other MTAs handle this. So
we make the CRs optional in all cases.

July 2003: Bare CRs cause trouble. We now treat them as line terminators as
well, so that there are no CRs in spooled messages. However, the message
terminating dot is not recognized between two bare CRs.

Arguments:
  fout      a FILE to which to write the message; NULL if skipping

Returns:    One of the END_xxx values indicating why it stopped reading
*/

static int
read_message_data_smtp(FILE *fout)
{
int ch_state = 0;
int ch;
int linelength = 0;

while ((ch = (receive_getc)(GETC_BUFFER_UNLIMITED)) != EOF)
  {
  if (ch == 0) body_zerocount++;
  switch (ch_state)
    {
    case 0:                             /* After LF or CRLF */
    if (ch == '.')
      {
      ch_state = 3;
      continue;                         /* Don't ever write . after LF */
      }
    ch_state = 1;

    /* Else fall through to handle as normal uschar. */

    case 1:                             /* Normal state */
    if (ch == '\n')
      {
      ch_state = 0;
      body_linecount++;
      if (linelength > max_received_linelength)
        max_received_linelength = linelength;
      linelength = -1;
      }
    else if (ch == '\r')
      {
      ch_state = 2;
      continue;
      }
    break;

    case 2:                             /* After (unwritten) CR */
    body_linecount++;
    if (linelength > max_received_linelength)
      max_received_linelength = linelength;
    linelength = -1;
    if (ch == '\n')
      {
      ch_state = 0;
      }
    else
      {
      message_size++;
      if (fout != NULL && fputc('\n', fout) == EOF) return END_WERROR;
      cutthrough_data_put_nl();
      if (ch != '\r') ch_state = 1; else continue;
      }
    break;

    case 3:                             /* After [CR] LF . */
    if (ch == '\n')
      return END_DOT;
    if (ch == '\r')
      {
      ch_state = 4;
      continue;
      }
    /* The dot was removed at state 3. For a doubled dot, here, reinstate
    it to cutthrough. The current ch, dot or not, is passed both to cutthrough
    and to file below. */
    if (ch == '.')
      {
      uschar c= ch;
      cutthrough_data_puts(&c, 1);
      }
    ch_state = 1;
    break;

    case 4:                             /* After [CR] LF . CR */
    if (ch == '\n') return END_DOT;
    message_size++;
    body_linecount++;
    if (fout != NULL && fputc('\n', fout) == EOF) return END_WERROR;
    cutthrough_data_put_nl();
    if (ch == '\r')
      {
      ch_state = 2;
      continue;
      }
    ch_state = 1;
    break;
    }

  /* Add the character to the spool file, unless skipping; then loop for the
  next. */

  message_size++;
  linelength++;
  if (fout)
    {
    if (fputc(ch, fout) == EOF) return END_WERROR;
    if (message_size > thismessage_size_limit) return END_SIZE;
    }
  if(ch == '\n')
    cutthrough_data_put_nl();
  else
    {
    uschar c = ch;
    cutthrough_data_puts(&c, 1);
    }
  }

/* Fall through here if EOF encountered. This indicates some kind of error,
since a correct message is terminated by [CR] LF . [CR] LF. */

return END_EOF;
}




/* Variant of the above read_message_data_smtp() specialised for RFC 3030
CHUNKING. Accept input lines separated by either CRLF or CR or LF and write
LF-delimited spoolfile.  Until we have wireformat spoolfiles, we need the
body_linecount accounting for proper re-expansion for the wire, so use
a cut-down version of the state-machine above; we don't need to do leading-dot
detection and unstuffing.

Arguments:
  fout      a FILE to which to write the message; NULL if skipping;
            must be open for both writing and reading.

Returns:    One of the END_xxx values indicating why it stopped reading
*/

static int
read_message_bdat_smtp(FILE * fout)
{
int linelength = 0, ch;
enum CH_STATE ch_state = LF_SEEN;
BOOL fix_nl = FALSE;

for(;;)
  {
  switch ((ch = bdat_getc(GETC_BUFFER_UNLIMITED)))
    {
    case EOF:   return END_EOF;
    case ERR:   return END_PROTOCOL;
    case EOD:
      /* Nothing to get from the sender anymore. We check the last
      character written to the spool.

      RFC 3030 states, that BDAT chunks are normal text, terminated by CRLF.
      If we would be strict, we would refuse such broken messages.
      But we are liberal, so we fix it.  It would be easy just to append
      the "\n" to the spool.

      But there are some more things (line counting, message size calculation and such),
      that would need to be duplicated here.  So we simply do some ungetc
      trickery.
      */
      if (fout)
        {
        if (fseek(fout, -1, SEEK_CUR) < 0)      return END_PROTOCOL;
        if (fgetc(fout) == '\n')                return END_DOT;
        }

      if (linelength == -1)    /* \r already seen (see below) */
        {
        DEBUG(D_receive) debug_printf("Add missing LF\n");
        bdat_ungetc('\n');
        continue;
        }
      DEBUG(D_receive) debug_printf("Add missing CRLF\n");
      bdat_ungetc('\r');      /* not even \r was seen */
      fix_nl = TRUE;

      continue;
    case '\0':  body_zerocount++; break;
    }
  switch (ch_state)
    {
    case LF_SEEN:                             /* After LF or CRLF */
      ch_state = MID_LINE;
      /* fall through to handle as normal uschar. */

    case MID_LINE:                            /* Mid-line state */
      if (ch == '\n')
        {
        ch_state = LF_SEEN;
        body_linecount++;
        if (linelength > max_received_linelength)
          max_received_linelength = linelength;
        linelength = -1;
        }
      else if (ch == '\r')
        {
        ch_state = CR_SEEN;
       if (fix_nl) bdat_ungetc('\n');
        continue;                       /* don't write CR */
        }
      break;

    case CR_SEEN:                       /* After (unwritten) CR */
      body_linecount++;
      if (linelength > max_received_linelength)
        max_received_linelength = linelength;
      linelength = -1;
      if (ch == '\n')
        ch_state = LF_SEEN;
      else
        {
        message_size++;
        if (fout && fputc('\n', fout) == EOF) return END_WERROR;
        cutthrough_data_put_nl();
        if (ch == '\r') continue;       /* don't write CR */
        ch_state = MID_LINE;
        }
      break;
    }

  /* Add the character to the spool file, unless skipping */

  message_size++;
  linelength++;
  if (fout)
    {
    if (fputc(ch, fout) == EOF) return END_WERROR;
    if (message_size > thismessage_size_limit) return END_SIZE;
    }
  if(ch == '\n')
    cutthrough_data_put_nl();
  else
    {
    uschar c = ch;
    cutthrough_data_puts(&c, 1);
    }
  }
/*NOTREACHED*/
}

static int
read_message_bdat_smtp_wire(FILE * fout)
{
int ch;

/* Remember that this message uses wireformat. */

DEBUG(D_receive) debug_printf("CHUNKING: %s\n",
        fout ? "writing spoolfile in wire format" : "flushing input");
f.spool_file_wireformat = TRUE;

for (;;)
  {
  if (chunking_data_left > 0)
    {
    unsigned len = MAX(chunking_data_left, thismessage_size_limit - message_size + 1);
    uschar * buf = bdat_getbuf(&len);

    if (!buf) return END_EOF;
    message_size += len;
    if (fout && fwrite(buf, len, 1, fout) != 1) return END_WERROR;
    }
  else switch (ch = bdat_getc(GETC_BUFFER_UNLIMITED))
    {
    case EOF: return END_EOF;
    case EOD: return END_DOT;
    case ERR: return END_PROTOCOL;

    default:
      message_size++;
  /*XXX not done:
  linelength
  max_received_linelength
  body_linecount
  body_zerocount
  */
      if (fout && fputc(ch, fout) == EOF) return END_WERROR;
      break;
    }
  if (message_size > thismessage_size_limit) return END_SIZE;
  }
/*NOTREACHED*/
}




/*************************************************
*             Swallow SMTP message               *
*************************************************/

/* This function is called when there has been some kind of error while reading
an SMTP message, and the remaining data may need to be swallowed. It is global
because it is called from smtp_closedown() to shut down an incoming call
tidily.

Argument:    a FILE from which to read the message
Returns:     nothing
*/

void
receive_swallow_smtp(void)
{
if (message_ended >= END_NOTENDED)
  message_ended = chunking_state <= CHUNKING_OFFERED
     ? read_message_data_smtp(NULL)
     : read_message_bdat_smtp_wire(NULL);
}



/*************************************************
*           Handle lost SMTP connection          *
*************************************************/

/* This function logs connection loss incidents and generates an appropriate
SMTP response.

Argument:  additional data for the message
Returns:   the SMTP response
*/

static uschar *
handle_lost_connection(uschar * s)
{
log_write(L_lost_incoming_connection | L_smtp_connection, LOG_MAIN,
  "%s lost while reading message data%s", smtp_get_connection_info(), s);
smtp_notquit_exit(US"connection-lost", NULL, NULL);
return US"421 Lost incoming connection";
}




/*************************************************
*         Handle a non-smtp reception error      *
*************************************************/

/* This function is called for various errors during the reception of non-SMTP
messages. It either sends a message to the sender of the problem message, or it
writes to the standard error stream.

Arguments:
  errcode     code for moan_to_sender(), identifying the error
  text1       first message text, passed to moan_to_sender()
  text2       second message text, used only for stderrr
  error_rc    code to pass to exim_exit if no problem
  f           FILE containing body of message (may be stdin)
  hptr        pointer to instore headers or NULL

Returns:      calls exim_exit(), which does not return
*/

static void
give_local_error(int errcode, uschar *text1, uschar *text2, int error_rc,
  FILE *f, header_line *hptr)
{
if (error_handling == ERRORS_SENDER)
  {
  error_block eblock;
  eblock.next = NULL;
  eblock.text1 = text1;
  eblock.text2 = US"";
  if (!moan_to_sender(errcode, &eblock, hptr, f, FALSE))
    error_rc = EXIT_FAILURE;
  }
else
  fprintf(stderr, "exim: %s%s\n", text2, text1);  /* Sic */
(void)fclose(f);
exim_exit(error_rc);
}



/*************************************************
*          Add header lines set up by ACL        *
*************************************************/

/* This function is called to add the header lines that were set up by
statements in an ACL to the list of headers in memory. It is done in two stages
like this, because when the ACL for RCPT is running, the other headers have not
yet been received. This function is called twice; once just before running the
DATA ACL, and once after. This is so that header lines added by MAIL or RCPT
are visible to the DATA ACL.

Originally these header lines were added at the end. Now there is support for
three different places: top, bottom, and after the Received: header(s). There
will always be at least one Received: header, even if it is marked deleted, and
even if something else has been put in front of it.

Arguments:
  acl_name   text to identify which ACL

Returns:     nothing
*/

static void
add_acl_headers(int where, uschar * acl_name)
{
header_line * last_received = NULL;

switch(where)
  {
  case ACL_WHERE_DKIM:
  case ACL_WHERE_MIME:
  case ACL_WHERE_DATA:
    if (  cutthrough.cctx.sock >= 0 && cutthrough.delivery
       && (acl_removed_headers || acl_added_headers))
    {
    log_write(0, LOG_MAIN|LOG_PANIC, "Header modification in data ACLs"
                        " will not take effect on cutthrough deliveries");
    return;
    }
  }

if (acl_removed_headers)
  {
  DEBUG(D_receive|D_acl) debug_printf_indent(">>Headers removed by %s ACL:\n", acl_name);

  for (header_line * h = header_list; h; h = h->next) if (h->type != htype_old)
    {
    const uschar * list = acl_removed_headers, * s;
    int sep = ':';         /* This is specified as a colon-separated list */

    /* If a list element has a leading '^' then it is an RE for
    the whole header, else just a header name. */
    while ((s = string_nextinlist(&list, &sep, NULL, 0)))
      if (  (  *s == '^'
            && regex_match(
                regex_must_compile(s, MCS_CACHEABLE, FALSE),
                h->text, h->slen, NULL)
            )
         || header_testname(h, s, Ustrlen(s), FALSE)
         )
        {
        h->type = htype_old;
        DEBUG(D_receive|D_acl) debug_printf_indent("  %s", h->text);
        }
    }
  acl_removed_headers = NULL;
  DEBUG(D_receive|D_acl) debug_printf_indent(">>\n");
  }

if (!acl_added_headers) return;
DEBUG(D_receive|D_acl) debug_printf_indent(">>Headers added by %s ACL:\n", acl_name);

for (header_line * h = acl_added_headers, * next; h; h = next)
  {
  next = h->next;

  switch(h->type)
    {
    case htype_add_top:
      h->next = header_list;
      header_list = h;
      DEBUG(D_receive|D_acl) debug_printf_indent("  (at top)");
      break;

    case htype_add_rec:
      if (!last_received)
        {
        last_received = header_list;
        while (!header_testname(last_received, US"Received", 8, FALSE))
          last_received = last_received->next;
        while (last_received->next &&
               header_testname(last_received->next, US"Received", 8, FALSE))
          last_received = last_received->next;
        }
      h->next = last_received->next;
      last_received->next = h;
      DEBUG(D_receive|D_acl) debug_printf_indent("  (after Received:)");
      break;

    case htype_add_rfc:
      /* add header before any header which is NOT Received: or Resent- */
      last_received = header_list;
      while ( last_received->next &&
              ( (header_testname(last_received->next, US"Received", 8, FALSE)) ||
                (header_testname_incomplete(last_received->next, US"Resent-", 7, FALSE)) ) )
                last_received = last_received->next;
      /* last_received now points to the last Received: or Resent-* header
         in an uninterrupted chain of those header types (seen from the beginning
         of all headers. Our current header must follow it. */
      h->next = last_received->next;
      last_received->next = h;
      DEBUG(D_receive|D_acl) debug_printf_indent("  (before any non-Received: or Resent-*: header)");
      break;

    default:
      h->next = NULL;
      header_last->next = h;
      DEBUG(D_receive|D_acl) debug_printf_indent("  ");
      break;
    }

  if (!h->next) header_last = h;

  /* Check for one of the known header types (From:, To:, etc.) though in
  practice most added headers are going to be "other". Lower case
  identification letters are never stored with the header; they are used
  for existence tests when messages are received. So discard any lower case
  flag values. */

  h->type = header_checkname(h, FALSE);
  if (h->type >= 'a') h->type = htype_other;

  DEBUG(D_receive|D_acl) debug_printf("%s", h->text);
  }

acl_added_headers = NULL;
DEBUG(D_receive|D_acl) debug_printf_indent(">>\n");
}



/*************************************************
*       Add host information for log line        *
*************************************************/

/* Called for acceptance and rejecting log lines. This adds information about
the calling host to a string that is being built dynamically.

Arguments:
  s           the dynamic string

Returns:      the extended string
*/

static gstring *
add_host_info_for_log(gstring * g)
{
if (sender_fullhost)
  {
  if (LOGGING(dnssec) && sender_host_dnssec)    /*XXX sender_helo_dnssec? */
    g = string_catn(g, US" DS", 3);
  g = string_append(g, 2, US" H=", sender_fullhost);
  if (LOGGING(incoming_interface) && interface_address)
    g = string_fmt_append(g, " I=[%s]:%d", interface_address, interface_port);
  }
if (f.tcp_in_fastopen && !f.tcp_in_fastopen_logged)
  {
  g = string_catn(g, US" TFO*", f.tcp_in_fastopen_data ? 5 : 4);
  f.tcp_in_fastopen_logged = TRUE;
  }
if (sender_ident)
  g = string_append(g, 2, US" U=", sender_ident);
if (LOGGING(connection_id))
  g = string_fmt_append(g, " Ci=%lu", connection_id);
if (received_protocol)
  g = string_append(g, 2, US" P=", received_protocol);
if (LOGGING(pipelining) && f.smtp_in_pipelining_advertised)
  {
  g = string_catn(g, US" L", 2);
#ifndef DISABLE_PIPE_CONNECT
  if (f.smtp_in_early_pipe_used)
    g = string_catn(g, US"*", 1);
  else if (f.smtp_in_early_pipe_advertised)
    g = string_catn(g, US".", 1);
#endif
  if (!f.smtp_in_pipelining_used)
    g = string_catn(g, US"-", 1);
  }
return g;
}



#ifdef WITH_CONTENT_SCAN

/*************************************************
*       Run the MIME ACL on a message            *
*************************************************/

/* This code is in a subroutine so that it can be used for both SMTP
and non-SMTP messages. It is called with a non-NULL ACL pointer.

Arguments:
  acl                The ACL to run (acl_smtp_mime or acl_not_smtp_mime)
  smtp_yield_ptr     Set FALSE to kill messages after dropped connection
  smtp_reply_ptr     Where SMTP reply is being built
  blackholed_by_ptr  Where "blackholed by" message is being built

Returns:             TRUE to carry on; FALSE to abandon the message
*/

static BOOL
run_mime_acl(uschar *acl, BOOL *smtp_yield_ptr, uschar **smtp_reply_ptr,
  uschar **blackholed_by_ptr)
{
FILE *mbox_file;
uschar * rfc822_file_path = NULL;
unsigned long mbox_size;
uschar *user_msg, *log_msg;
int mime_part_count_buffer = -1;
uschar * mbox_filename;
int rc = OK;

/* check if it is a MIME message */

for (header_line * my_headerlist = header_list; my_headerlist;
     my_headerlist = my_headerlist->next)
  if (  my_headerlist->type != '*'                      /* skip deleted headers */
     && strncmpic(my_headerlist->text, US"Content-Type:", 13) == 0
     )
    {
    DEBUG(D_receive) debug_printf("Found Content-Type: header - executing acl_smtp_mime.\n");
    goto DO_MIME_ACL;
    }

DEBUG(D_receive) debug_printf("No Content-Type: header - presumably not a MIME message.\n");
return TRUE;

DO_MIME_ACL:

/* make sure the eml mbox file is spooled up */
if (!(mbox_file = spool_mbox(&mbox_size, NULL, &mbox_filename)))
  {                                                             /* error while spooling */
  log_write(0, LOG_MAIN|LOG_PANIC,
         "acl_smtp_mime: error while creating mbox spool file, message temporarily rejected.");
  Uunlink(spool_name);
  unspool_mbox();
#ifdef EXPERIMENTAL_DCC
  dcc_ok = 0;
#endif
  smtp_respond(US"451", 3, SR_FINAL, US"temporary local problem");
  message_id[0] = 0;            /* Indicate no message accepted */
  *smtp_reply_ptr = US"";       /* Indicate reply already sent */
  return FALSE;                 /* Indicate skip to end of receive function */
  }

mime_is_rfc822 = 0;

MIME_ACL_CHECK:
mime_part_count = -1;
rc = mime_acl_check(acl, mbox_file, NULL, &user_msg, &log_msg);
(void)fclose(mbox_file);

if (rfc822_file_path)
  {
  mime_part_count = mime_part_count_buffer;

  if (unlink(CS rfc822_file_path) == -1)
    {
    log_write(0, LOG_PANIC,
         "acl_smtp_mime: can't unlink RFC822 spool file, skipping.");
    goto END_MIME_ACL;
    }
  rfc822_file_path = NULL;
  }

/* check if we must check any message/rfc822 attachments */
if (rc == OK)
  {
  uschar * scandir = string_copyn(mbox_filename,
              Ustrrchr(mbox_filename, '/') - mbox_filename);
  struct dirent * entry;
  DIR * tempdir;

  for (tempdir = exim_opendir(scandir); entry = readdir(tempdir); )
    if (strncmpic(US entry->d_name, US"__rfc822_", 9) == 0)
      {
      rfc822_file_path = string_sprintf("%s/%s", scandir, entry->d_name);
      DEBUG(D_receive)
        debug_printf("RFC822 attachment detected: running MIME ACL for '%s'\n",
          rfc822_file_path);
      break;
      }
  closedir(tempdir);

  if (rfc822_file_path)
    {
    if ((mbox_file = Ufopen(rfc822_file_path, "rb")))
      {
      /* set RFC822 expansion variable */
      mime_is_rfc822 = 1;
      mime_part_count_buffer = mime_part_count;
      goto MIME_ACL_CHECK;
      }
    log_write(0, LOG_PANIC,
       "acl_smtp_mime: can't open RFC822 spool file, skipping.");
    unlink(CS rfc822_file_path);
    }
  }

END_MIME_ACL:
add_acl_headers(ACL_WHERE_MIME, US"MIME");
if (rc == DISCARD)
  {
  recipients_count = 0;
  *blackholed_by_ptr = US"MIME ACL";
  cancel_cutthrough_connection(TRUE, US"mime acl discard");
  }
else if (rc != OK)
  {
  Uunlink(spool_name);
  cancel_cutthrough_connection(TRUE, US"mime acl not ok");
  unspool_mbox();
#ifdef EXPERIMENTAL_DCC
  dcc_ok = 0;
#endif
  if (smtp_input)
    {
    if (smtp_handle_acl_fail(ACL_WHERE_MIME, rc, user_msg, log_msg) != 0)
      *smtp_yield_ptr = FALSE;  /* No more messages after dropped connection */
    *smtp_reply_ptr = US"";     /* Indicate reply already sent */
    }
  message_id[0] = 0;            /* Indicate no message accepted */
  return FALSE;                 /* Cause skip to end of receive function */
  }

return TRUE;
}

#endif  /* WITH_CONTENT_SCAN */



void
received_header_gen(void)
{
uschar * received;
uschar * timestamp = expand_string(US"${tod_full}");
header_line * received_header= header_list;

if (recipients_count == 1) received_for = recipients_list[0].address;
received = expand_string(received_header_text);
received_for = NULL;

if (!received)
  {
  if(spool_name[0] != 0)
    Uunlink(spool_name);           /* Lose the data file */
  log_write(0, LOG_MAIN|LOG_PANIC_DIE, "Expansion of \"%s\" "
    "(received_header_text) failed: %s", string_printing(received_header_text),
      expand_string_message);
  }

/* The first element on the header chain is reserved for the Received header,
so all we have to do is fill in the text pointer, and set the type. However, if
the result of the expansion is an empty string, we leave the header marked as
"old" so as to refrain from adding a Received header. */

if (!received[0])
  {
  received_header->text = string_sprintf("Received: ; %s\n", timestamp);
  received_header->type = htype_old;
  }
else
  {
  received_header->text = string_sprintf("%s;\n\t%s\n", received, timestamp);
  received_header->type = htype_received;
  }

received_header->slen = Ustrlen(received_header->text);

DEBUG(D_receive) debug_printf(">>Generated Received: header line\n%c %s",
  received_header->type, received_header->text);
}



/*************************************************
*                 Receive message                *
*************************************************/

/* Receive a message on the given input, and put it into a pair of spool files.
Either a non-null list of recipients, or the extract flag will be true, or
both. The flag sender_local is true for locally generated messages. The flag
submission_mode is true if an ACL has obeyed "control = submission". The flag
suppress_local_fixups is true if an ACL has obeyed "control =
suppress_local_fixups" or -G was passed on the command-line.
The flag smtp_input is true if the message is to be
handled using SMTP conventions about termination and lines starting with dots.
For non-SMTP messages, dot_ends is true for dot-terminated messages.

If a message was successfully read, message_id[0] will be non-zero.

The general actions of this function are:

  . Read the headers of the message (if any) into a chain of store
    blocks.

  . If there is a "sender:" header and the message is locally originated,
    throw it away, unless the caller is trusted, or unless
    active_local_sender_retain is set - which can only happen if
    active_local_from_check is false.

  . If recipients are to be extracted from the message, build the
    recipients list from the headers, removing any that were on the
    original recipients list (unless extract_addresses_remove_arguments is
    false), and at the same time, remove any bcc header that may be present.

  . Get the spool file for the data, sort out its unique name, open
    and lock it (but don't give it the name yet).

  . Generate a "Message-Id" header if the message doesn't have one, for
    locally-originated messages.

  . Generate a "Received" header.

  . Ensure the recipients list is fully qualified and rewritten if necessary.

  . If there are any rewriting rules, apply them to the sender address
    and also to the headers.

  . If there is no from: header, generate one, for locally-generated messages
    and messages in "submission mode" only.

  . If the sender is local, check that from: is correct, and if not, generate
    a Sender: header, unless message comes from a trusted caller, or this
    feature is disabled by active_local_from_check being false.

  . If there is no "date" header, generate one, for locally-originated
    or submission mode messages only.

  . Copy the rest of the input, or up to a terminating "." if in SMTP or
    dot_ends mode, to the data file. Leave it open, to hold the lock.

  . Write the envelope and the headers to a new file.

  . Set the name for the header file; close it.

  . Set the name for the data file; close it.

Because this function can potentially be called many times in a single
SMTP connection, all store should be got by store_get(), so that it will be
automatically retrieved after the message is accepted.

FUDGE: It seems that sites on the net send out messages with just LF
terminators, despite the warnings in the RFCs, and other MTAs handle this. So
we make the CRs optional in all cases.

July 2003: Bare CRs in messages, especially in header lines, cause trouble. A
new regime is now in place in which bare CRs in header lines are turned into LF
followed by a space, so as not to terminate the header line.

February 2004: A bare LF in a header line in a message whose first line was
terminated by CRLF is treated in the same way as a bare CR.

Arguments:
  extract_recip  TRUE if recipients are to be extracted from the message's
                   headers

Returns:  TRUE   there are more messages to be read (SMTP input)
          FALSE  there are no more messages to be read (non-SMTP input
                 or SMTP connection collapsed, or other failure)

When reading a message for filter testing, the returned value indicates
whether the headers (which is all that is read) were terminated by '.' or
not. */

BOOL
receive_msg(BOOL extract_recip)
{
int  rc = FAIL;
int  msg_size = 0;
int  process_info_len = Ustrlen(process_info);
int  error_rc = error_handling == ERRORS_SENDER
        ? errors_sender_rc : EXIT_FAILURE;
int  header_size = 256;
int  had_zero = 0;
int  prevlines_length = 0;
const int id_resolution = BASE_62 == 62 && !host_number_string ? 1
  : BASE_62 != 62 && host_number_string ? 4
  : 2;

int ptr = 0;

BOOL contains_resent_headers = FALSE;
BOOL extracted_ignored = FALSE;
BOOL first_line_ended_crlf = TRUE_UNSET;
BOOL smtp_yield = TRUE;
BOOL yield = FALSE;

BOOL resents_exist = FALSE;
uschar *resent_prefix = US"";
uschar *blackholed_by = NULL;
uschar *blackhole_log_msg = US"";
enum {NOT_TRIED, TMP_REJ, PERM_REJ, ACCEPTED} cutthrough_done = NOT_TRIED;

flock_t lock_data;
error_block *bad_addresses = NULL;

uschar *frozen_by = NULL;
uschar *queued_by = NULL;

uschar *errmsg;
rmark rcvd_log_reset_point;
gstring * g;
struct stat statbuf;

/* Final message to give to SMTP caller, and messages from ACLs */

uschar *smtp_reply = NULL;
uschar *user_msg, *log_msg;

/* Working header pointers */

rmark reset_point;
header_line *next;

/* Flags for noting the existence of certain headers (only one left) */

BOOL date_header_exists = FALSE;

/* Pointers to receive the addresses of headers whose contents we need. */

header_line *from_header = NULL;
header_line *subject_header = NULL;
header_line *msgid_header = NULL;
header_line *received_header;
BOOL msgid_header_newly_created = FALSE;

/* Variables for use when building the Received: header. */

uschar *timestamp;
int tslen;

/* Time of creation of message_id */

static struct timeval message_id_tv = { 0, 0 };


/* Release any open files that might have been cached while preparing to
accept the message - e.g. by verifying addresses - because reading a message
might take a fair bit of real time. */

search_tidyup();

/* Extracting the recipient list from an input file is incompatible with
cutthrough delivery with the no-spool option.  It shouldn't be possible
to set up the combination, but just in case kill any ongoing connection. */
if (extract_recip || !smtp_input)
  cancel_cutthrough_connection(TRUE, US"not smtp input");

/* Initialize the chain of headers by setting up a place-holder for Received:
header. Temporarily mark it as "old", i.e. not to be used. We keep header_last
pointing to the end of the chain to make adding headers simple. */

received_header = header_list = header_last = store_get(sizeof(header_line), GET_UNTAINTED);
header_list->next = NULL;
header_list->type = htype_old;
header_list->text = NULL;
header_list->slen = 0;

/* Control block for the next header to be read.
The data comes from the message, so is tainted. */

reset_point = store_mark();
next = store_get(sizeof(header_line), GET_UNTAINTED);
next->text = store_get(header_size, GET_TAINTED);

/* Initialize message id to be null (indicating no message read), and the
header names list to be the normal list. Indicate there is no data file open
yet, initialize the size and warning count, and deal with no size limit. */

message_id[0] = 0;
spool_data_file = NULL;
data_fd = -1;
spool_name = US"";
message_size = 0;
warning_count = 0;
received_count = 1;            /* For the one we will add */

if (thismessage_size_limit <= 0) thismessage_size_limit = INT_MAX;

/* While reading the message, the following counts are computed. */

message_linecount = body_linecount = body_zerocount =
  max_received_linelength = 0;

#ifdef WITH_CONTENT_SCAN
/* reset non-per-part mime variables */
mime_is_coverletter    = 0;
mime_is_rfc822         = 0;
mime_part_count        = -1;
#endif

#ifndef DISABLE_DKIM
/* Call into DKIM to set up the context.  In CHUNKING mode
we clear the dot-stuffing flag */
if (smtp_input && !smtp_batched_input && !f.dkim_disable_verify)
  dkim_exim_verify_init(chunking_state <= CHUNKING_OFFERED);
#endif

#ifdef SUPPORT_DMARC
if (sender_host_address) dmarc_init();  /* initialize libopendmarc */
#endif

/* In SMTP sessions we may receive several messages in one connection. Before
each subsequent one, we wait for the clock to tick at the level of message-id
granularity.
This is so that the combination of time+pid is unique, even on systems where the
pid can be re-used within our time interval. We can't shorten the interval
without re-designing the message-id. See comments above where the message id is
created. This is Something For The Future.
Do this wait any time we have previously created a message-id, even if we
rejected the message.  This gives unique IDs for logging done by ACLs.
The initial timestamp must have been obtained via exim_gettime() to avoid
issues on Linux with suspend/resume. */

if (message_id_tv.tv_sec)
  {
  message_id_tv.tv_usec = (message_id_tv.tv_usec/id_resolution) * id_resolution;
  exim_wait_tick(&message_id_tv, id_resolution);
  }

/* Remember the time of reception. Exim uses time+pid for uniqueness of message
ids, and fractions of a second are required. See the comments that precede the
message id creation below.
We use a routine that if possible uses a monotonic clock, and can be used again
after reception for the tick-wait even under the Linux non-Posix behaviour. */

else
  exim_gettime(&message_id_tv);

/* For other uses of the received time we can operate with granularity of one
second, and for that we use the global variable received_time. This is for
things like ultimate message timeouts.
For this we do not care about the Linux suspend/resume problem, so rather than
use exim_gettime() everywhere we use a plain gettimeofday() here. */

gettimeofday(&received_time, NULL);

/* If SMTP input, set the special handler for timeouts. The alarm() calls
happen in the smtp_getc() function when it refills its buffer. */

had_data_timeout = 0;
if (smtp_input)
  os_non_restarting_signal(SIGALRM, data_timeout_handler);

/* If not SMTP input, timeout happens only if configured, and we just set a
single timeout for the whole message. */

else if (receive_timeout > 0)
  {
  os_non_restarting_signal(SIGALRM, data_timeout_handler);
  ALARM(receive_timeout);
  }

/* SIGTERM and SIGINT are caught always. */

had_data_sigint = 0;
signal(SIGTERM, data_sigterm_sigint_handler);
signal(SIGINT, data_sigterm_sigint_handler);

/* Header lines in messages are not supposed to be very long, though when
unfolded, to: and cc: headers can take up a lot of store. We must also cope
with the possibility of junk being thrown at us. Start by getting 256 bytes for
storing the header, and extend this as necessary using string_cat().

To cope with total lunacies, impose an upper limit on the length of the header
section of the message, as otherwise the store will fill up. We must also cope
with the possibility of binary zeros in the data. Hence we cannot use fgets().
Folded header lines are joined into one string, leaving the '\n' characters
inside them, so that writing them out reproduces the input.

Loop for each character of each header; the next structure for chaining the
header is set up already, with ptr the offset of the next character in
next->text. */

for (;;)
  {
  int ch = (receive_getc)(GETC_BUFFER_UNLIMITED);

  /* If we hit EOF on a SMTP connection, it's an error, since incoming
  SMTP must have a correct "." terminator. */

  if (smtp_input /* && !smtp_batched_input */)
    if (ch == EOF)
      {
      smtp_reply = handle_lost_connection(US" (header)");
      smtp_yield = FALSE;
      goto TIDYUP;                       /* Skip to end of function */
      }
    else if (ch == ERR)
      goto TIDYUP;

  /* See if we are at the current header's size limit - there must be at least
  four bytes left. This allows for the new character plus a zero, plus two for
  extra insertions when we are playing games with dots and carriage returns. If
  we are at the limit, extend the text buffer. This could have been done
  automatically using string_cat() but because this is a tightish loop storing
  only one character at a time, we choose to do it inline. Normally
  store_extend() will be able to extend the block; only at the end of a big
  store block will a copy be needed. To handle the case of very long headers
  (and sometimes lunatic messages can have ones that are 100s of K long) we
  call store_release() for strings that have been copied - if the string is at
  the start of a block (and therefore the only thing in it, because we aren't
  doing any other gets), the block gets freed. We can only do this release if
  there were no allocations since the once that we want to free. */

  if (ptr >= header_size - 4)
    {
    int oldsize = header_size;

    if (header_size >= INT_MAX/2)
      goto OVERSIZE;
    header_size *= 2;

    if (!store_extend(next->text, oldsize, header_size))
      next->text = store_newblock(next->text, header_size, ptr);
    }

  /* Cope with receiving a binary zero. There is dispute about whether
  these should be allowed in RFC 822 messages. The middle view is that they
  should not be allowed in headers, at least. Exim takes this attitude at
  the moment. We can't just stomp on them here, because we don't know that
  this line is a header yet. Set a flag to cause scanning later. */

  if (ch == 0) had_zero++;

  /* Test for termination. Lines in remote SMTP are terminated by CRLF, while
  those from data files use just LF. Treat LF in local SMTP input as a
  terminator too. Treat EOF as a line terminator always. */

  if (ch < 0) goto EOL;

  /* FUDGE: There are sites out there that don't send CRs before their LFs, and
  other MTAs accept this. We are therefore forced into this "liberalisation"
  too, so we accept LF as a line terminator whatever the source of the message.
  However, if the first line of the message ended with a CRLF, we treat a bare
  LF specially by inserting a white space after it to ensure that the header
  line is not terminated. */

  if (ch == '\n')
    {
    if (first_line_ended_crlf == TRUE_UNSET)
      first_line_ended_crlf = FALSE;
    else if (first_line_ended_crlf)
      receive_ungetc(' ');
    goto EOL;
    }

  /* This is not the end of the line. If this is SMTP input and this is
  the first character in the line and it is a "." character, ignore it.
  This implements the dot-doubling rule, though header lines starting with
  dots aren't exactly common. They are legal in RFC 822, though. If the
  following is CRLF or LF, this is the line that that terminates the

  entire message. We set message_ended to indicate this has happened (to
  prevent further reading), and break out of the loop, having freed the
  empty header, and set next = NULL to indicate no data line. */

  if (f.dot_ends && ptr == 0 && ch == '.')
    {
    ch = (receive_getc)(GETC_BUFFER_UNLIMITED);
    if (ch == '\n' && first_line_ended_crlf == TRUE /* and not TRUE_UNSET */ )
                /* dot, LF  but we are in CRLF mode.  Attack? */
      ch = ' '; /* replace the LF with a space */

    else if (ch == '\r')
      {
      ch = (receive_getc)(GETC_BUFFER_UNLIMITED);
      if (ch != '\n')
        {
        if (ch >= 0) receive_ungetc(ch);
        ch = '\r';              /* Revert to CR */
        }
      }
    if (ch == '\n')
      {
      message_ended = END_DOT;
      reset_point = store_reset(reset_point);
      next = NULL;
      break;                    /* End character-reading loop */
      }

    /* For non-SMTP input, the dot at the start of the line was really a data
    character. What is now in ch is the following character. We guaranteed
    enough space for this above. */

    if (!smtp_input)
      {
      next->text[ptr++] = '.';
      message_size++;
      }
    }

  /* If CR is immediately followed by LF, end the line, ignoring the CR, and
  remember this case if this is the first line ending. */

  if (ch == '\r')
    {
    ch = (receive_getc)(GETC_BUFFER_UNLIMITED);
    if (ch == '\n')
      {
      if (first_line_ended_crlf == TRUE_UNSET)
        first_line_ended_crlf = TRUE;
      goto EOL;
      }

    /* Otherwise, put back the character after CR, and turn the bare CR
    into LF SP. */

    if (ch >= 0) (receive_ungetc)(ch);
    next->text[ptr++] = '\n';
    message_size++;
    ch = ' ';
    }

  /* We have a data character for the header line. */

  next->text[ptr++] = ch;    /* Add to buffer */
  message_size++;            /* Total message size so far */

  /* Handle failure due to a humungously long header section. The >= allows
  for the terminating \n. Add what we have so far onto the headers list so
  that it gets reflected in any error message, and back up the just-read
  character. */

  if (message_size >= header_maxsize)
    {
OVERSIZE:
    next->text[ptr] = 0;
    next->slen = ptr;
    next->type = htype_other;
    next->next = NULL;
    header_last->next = next;
    header_last = next;

    log_write(0, LOG_MAIN, "ridiculously long message header received from "
      "%s (more than %d characters): message abandoned",
      f.sender_host_unknown ? sender_ident : sender_fullhost, header_maxsize);

    if (smtp_input)
      {
      smtp_reply = US"552 Message header is ridiculously long";
      receive_swallow_smtp();
      goto TIDYUP;                             /* Skip to end of function */
      }

    else
      {
      give_local_error(ERRMESS_VLONGHEADER,
        string_sprintf("message header longer than %d characters received: "
         "message not accepted", header_maxsize), US"", error_rc, stdin,
           header_list->next);
      /* Does not return */
      }
    }

  continue;                  /* With next input character */

  /* End of header line reached */

  EOL:

  /* Keep track of lines for BSMTP errors and overall message_linecount. */

  receive_linecount++;
  message_linecount++;

  /* Keep track of maximum line length */

  if (ptr - prevlines_length > max_received_linelength)
    max_received_linelength = ptr - prevlines_length;
  prevlines_length = ptr + 1;

  /* Now put in the terminating newline. There is always space for
  at least two more characters. */

  next->text[ptr++] = '\n';
  message_size++;

  /* A blank line signals the end of the headers; release the unwanted
  space and set next to NULL to indicate this. */

  if (ptr == 1)
    {
    reset_point = store_reset(reset_point);
    next = NULL;
    break;
    }

  /* There is data in the line; see if the next input character is a
  whitespace character. If it is, we have a continuation of this header line.
  There is always space for at least one character at this point. */

  if (ch >= 0)
    {
    int nextch = (receive_getc)(GETC_BUFFER_UNLIMITED);
    if (nextch == ' ' || nextch == '\t')
      {
      next->text[ptr++] = nextch;
      if (++message_size >= header_maxsize)
        goto OVERSIZE;
      continue;                      /* Iterate the loop */
      }
    else if (nextch >= 0)       /* not EOF, ERR etc */
      (receive_ungetc)(nextch);   /* For next time */
    else ch = nextch;                   /* Cause main loop to exit at end */
    }

  /* We have got to the real line end. Terminate the string and release store
  beyond it. If it turns out to be a real header, internal binary zeros will
  be squashed later. */

  next->text[ptr] = 0;
  next->slen = ptr;
  store_release_above(next->text + ptr + 1);

  /* Check the running total size against the overall message size limit. We
  don't expect to fail here, but if the overall limit is set less than MESSAGE_
  MAXSIZE and a big header is sent, we want to catch it. Just stop reading
  headers - the code to read the body will then also hit the buffer. */

  if (message_size > thismessage_size_limit) break;

  /* A line that is not syntactically correct for a header also marks
  the end of the headers. In this case, we leave next containing the
  first data line. This might actually be several lines because of the
  continuation logic applied above, but that doesn't matter.

  It turns out that smail, and presumably sendmail, accept leading lines
  of the form

  From ph10 Fri Jan  5 12:35 GMT 1996

  in messages. The "mail" command on Solaris 2 sends such lines. I cannot
  find any documentation of this, but for compatibility it had better be
  accepted. Exim restricts it to the case of non-smtp messages, and
  treats it as an alternative to the -f command line option. Thus it is
  ignored except for trusted users or filter testing. Otherwise it is taken
  as the sender address, unless -f was used (sendmail compatibility).

  It further turns out that some UUCPs generate the From_line in a different
  format, e.g.

  From ph10 Fri, 7 Jan 97 14:00:00 GMT

  The regex for matching these things is now capable of recognizing both
  formats (including 2- and 4-digit years in the latter). In fact, the regex
  is now configurable, as is the expansion string to fish out the sender.

  Even further on it has been discovered that some broken clients send
  these lines in SMTP messages. There is now an option to ignore them from
  specified hosts or networks. Sigh. */

  if (  header_last == header_list
     && (  !smtp_input
        || (  sender_host_address
           && verify_check_host(&ignore_fromline_hosts) == OK
           )
        || (!sender_host_address && ignore_fromline_local)
        )
     && regex_match_and_setup(regex_From, next->text, 0, -1)
     )
    {
    if (!f.sender_address_forced)
      {
      uschar *uucp_sender = expand_string(uucp_from_sender);
      if (!uucp_sender)
        log_write(0, LOG_MAIN|LOG_PANIC,
          "expansion of \"%s\" failed after matching "
          "\"From \" line: %s", uucp_from_sender, expand_string_message);
      else
        {
        int start, end, domain;
        uschar *errmess;
        uschar *newsender = parse_extract_address(uucp_sender, &errmess,
          &start, &end, &domain, TRUE);
        if (newsender)
          {
          if (domain == 0 && newsender[0] != 0)
            /* deconst ok as newsender was not const */
            newsender = US rewrite_address_qualify(newsender, FALSE);

          if (filter_test != FTEST_NONE || receive_check_set_sender(newsender))
            {
            sender_address = newsender;

            if (f.trusted_caller || filter_test != FTEST_NONE)
              {
              authenticated_sender = NULL;
              originator_name = US"";
              f.sender_local = FALSE;
              }

            if (filter_test != FTEST_NONE)
              printf("Sender taken from \"From \" line\n");
            }
          }
        }
      }
    }

  /* Not a leading "From " line. Check to see if it is a valid header line.
  Header names may contain any non-control characters except space and colon,
  amazingly. */

  else
    {
    uschar * p = next->text;

    /* If not a valid header line, break from the header reading loop, leaving
    next != NULL, indicating that it holds the first line of the body. */

    if (isspace(*p)) break;
    while (mac_isgraph(*p) && *p != ':') p++;
    while (isspace(*p)) p++;
    if (*p != ':')
      {
      body_zerocount = had_zero;
      break;
      }

    /* We have a valid header line. If there were any binary zeroes in
    the line, stomp on them here. */

    if (had_zero > 0)
      for (uschar * p = next->text; p < next->text + ptr; p++) if (*p == 0)
        *p = '?';

    /* It is perfectly legal to have an empty continuation line
    at the end of a header, but it is confusing to humans
    looking at such messages, since it looks like a blank line.
    Reduce confusion by removing redundant white space at the
    end. We know that there is at least one printing character
    (the ':' tested for above) so there is no danger of running
    off the end. */

    p = next->text + ptr - 2;
    for (;;)
      {
      while (*p == ' ' || *p == '\t') p--;
      if (*p != '\n') break;
      ptr = (p--) - next->text + 1;
      message_size -= next->slen - ptr;
      next->text[ptr] = 0;
      next->slen = ptr;
      }

    /* Add the header to the chain */

    next->type = htype_other;
    next->next = NULL;
    header_last->next = next;
    header_last = next;

    /* Check the limit for individual line lengths. This comes after adding to
    the chain so that the failing line is reflected if a bounce is generated
    (for a local message). */

    if (header_line_maxsize > 0 && next->slen > header_line_maxsize)
      {
      log_write(0, LOG_MAIN, "overlong message header line received from "
        "%s (more than %d characters): message abandoned",
        f.sender_host_unknown ? sender_ident : sender_fullhost,
        header_line_maxsize);

      if (smtp_input)
        {
        smtp_reply = US"552 A message header line is too long";
        receive_swallow_smtp();
        goto TIDYUP;                             /* Skip to end of function */
        }

      else
        give_local_error(ERRMESS_VLONGHDRLINE,
          string_sprintf("message header line longer than %d characters "
           "received: message not accepted", header_line_maxsize), US"",
           error_rc, stdin, header_list->next);
        /* Does not return */
      }

    /* Note if any resent- fields exist. */

    if (!resents_exist && strncmpic(next->text, US"resent-", 7) == 0)
      {
      resents_exist = TRUE;
      resent_prefix = US"Resent-";
      }
    }

  /* Reject CHUNKING messages that do not CRLF their first header line */

  if (!first_line_ended_crlf && chunking_state > CHUNKING_OFFERED)
    {
    log_write(L_size_reject, LOG_MAIN|LOG_REJECT, "rejected from <%s>%s%s%s%s: "
      "Non-CRLF-terminated header, under CHUNKING: message abandoned",
      sender_address,
      sender_fullhost ? " H=" : "", sender_fullhost ? sender_fullhost : US"",
      sender_ident ? " U=" : "",    sender_ident ? sender_ident : US"");
    smtp_printf("552 Message header not CRLF terminated\r\n", SP_NO_MORE);
    bdat_flush_data();
    smtp_reply = US"";
    goto TIDYUP;                             /* Skip to end of function */
    }

  /* The line has been handled. If we have hit EOF, break out of the loop,
  indicating no pending data line and no more data for the message */

  if (ch < 0)
    {
    next = NULL;
    if (ch == EOF)      message_ended = END_DOT;
    else if (ch == ERR) message_ended = END_PROTOCOL;
    break;
    }

  /* Set up for the next header */

  reset_point = store_mark();
  header_size = 256;
  next = store_get(sizeof(header_line), GET_UNTAINTED);
  next->text = store_get(header_size, GET_TAINTED);
  ptr = 0;
  had_zero = 0;
  prevlines_length = 0;
  }      /* Continue, starting to read the next header */

/* At this point, we have read all the headers into a data structure in main
store. The first header is still the dummy placeholder for the Received: header
we are going to generate a bit later on. If next != NULL, it contains the first
data line - which terminated the headers before reaching a blank line (not the
normal case). */

DEBUG(D_receive)
  {
  debug_printf(">>Headers received:\n");
  for (header_line * h = header_list->next; h; h = h->next)
    debug_printf("%s", h->text);
  debug_printf("\n");
  }

/* End of file on any SMTP connection is an error. If an incoming SMTP call
is dropped immediately after valid headers, the next thing we will see is EOF.
We must test for this specially, as further down the reading of the data is
skipped if already at EOF.
In CHUNKING mode, a protocol error makes us give up on the message. */

if (smtp_input)
  if ((receive_feof)())
    {
    smtp_reply = handle_lost_connection(US" (after header)");
    smtp_yield = FALSE;
    goto TIDYUP;                        /* Skip to end of function */
    }
  else if (message_ended == END_PROTOCOL)
    {
    smtp_reply = US"";                  /* no reply needed */
    goto TIDYUP;
    }

/* If this is a filter test run and no headers were read, output a warning
in case there is a mistake in the test message. */

if (filter_test != FTEST_NONE && header_list->next == NULL)
  printf("Warning: no message headers read\n");


/* Scan the headers to identify them. Some are merely marked for later
processing; some are dealt with here. */

for (header_line * h = header_list->next; h; h = h->next)
  {
  BOOL is_resent = strncmpic(h->text, US"resent-", 7) == 0;
  if (is_resent) contains_resent_headers = TRUE;

  switch (header_checkname(h, is_resent))
    {
    case htype_bcc:
      h->type = htype_bcc;        /* Both Bcc: and Resent-Bcc: */
      break;

    case htype_cc:
      h->type = htype_cc;         /* Both Cc: and Resent-Cc: */
      break;

      /* Record whether a Date: or Resent-Date: header exists, as appropriate. */

    case htype_date:
      if (!resents_exist || is_resent) date_header_exists = TRUE;
      break;

      /* Same comments as about Return-Path: below. */

    case htype_delivery_date:
      if (delivery_date_remove) h->type = htype_old;
      break;

      /* Same comments as about Return-Path: below. */

    case htype_envelope_to:
      if (envelope_to_remove) h->type = htype_old;
      break;

      /* Mark all "From:" headers so they get rewritten. Save the one that is to
      be used for Sender: checking. For Sendmail compatibility, if the "From:"
      header consists of just the login id of the user who called Exim, rewrite
      it with the gecos field first. Apply this rule to Resent-From: if there
      are resent- fields. */

    case htype_from:
      h->type = htype_from;
      if (!resents_exist || is_resent)
        {
        from_header = h;
        if (!smtp_input)
          {
          int len;
          uschar *s = Ustrchr(h->text, ':') + 1;
          while (isspace(*s)) s++;
          len = h->slen - (s - h->text) - 1;
          if (Ustrlen(originator_login) == len &&
              strncmpic(s, originator_login, len) == 0)
            {
            uschar *name = is_resent? US"Resent-From" : US"From";
            header_add(htype_from, "%s: %s <%s@%s>\n", name, originator_name,
              originator_login, qualify_domain_sender);
            from_header = header_last;
            h->type = htype_old;
            DEBUG(D_receive|D_rewrite)
              debug_printf("rewrote \"%s:\" header using gecos\n", name);
           }
          }
        }
      break;

      /* Identify the Message-id: header for generating "in-reply-to" in the
      autoreply transport. For incoming logging, save any resent- value. In both
      cases, take just the first of any multiples. */

    case htype_id:
      if (!msgid_header && (!resents_exist || is_resent))
        {
        msgid_header = h;
        h->type = htype_id;
        }
      break;

      /* Flag all Received: headers */

    case htype_received:
      h->type = htype_received;
      received_count++;
      break;

      /* "Reply-to:" is just noted (there is no resent-reply-to field) */

    case htype_reply_to:
      h->type = htype_reply_to;
      break;

      /* The Return-path: header is supposed to be added to messages when
      they leave the SMTP system. We shouldn't receive messages that already
      contain Return-path. However, since Exim generates Return-path: on
      local delivery, resent messages may well contain it. We therefore
      provide an option (which defaults on) to remove any Return-path: headers
      on input. Removal actually means flagging as "old", which prevents the
      header being transmitted with the message. */

    case htype_return_path:
      if (return_path_remove) h->type = htype_old;

      /* If we are testing a mail filter file, use the value of the
      Return-Path: header to set up the return_path variable, which is not
      otherwise set. However, remove any <> that surround the address
      because the variable doesn't have these. */

      if (filter_test != FTEST_NONE)
        {
        uschar *start = h->text + 12;
        uschar *end = start + Ustrlen(start);
        while (isspace(*start)) start++;
        while (end > start && isspace(end[-1])) end--;
        if (*start == '<' && end[-1] == '>')
          {
          start++;
          end--;
          }
        return_path = string_copyn(start, end - start);
        printf("Return-path taken from \"Return-path:\" header line\n");
        }
      break;

    /* If there is a "Sender:" header and the message is locally originated,
    and from an untrusted caller and suppress_local_fixups is not set, or if we
    are in submission mode for a remote message, mark it "old" so that it will
    not be transmitted with the message, unless active_local_sender_retain is
    set. (This can only be true if active_local_from_check is false.) If there
    are any resent- headers in the message, apply this rule to Resent-Sender:
    instead of Sender:. Messages with multiple resent- header sets cannot be
    tidily handled. (For this reason, at least one MUA - Pine - turns old
    resent- headers into X-resent- headers when resending, leaving just one
    set.) */

    case htype_sender:
      h->type =    !f.active_local_sender_retain
                && (  f.sender_local && !f.trusted_caller && !f.suppress_local_fixups
                   || f.submission_mode
                   )
                && (!resents_exist || is_resent)
        ? htype_old : htype_sender;
      break;

      /* Remember the Subject: header for logging. There is no Resent-Subject */

    case htype_subject:
      subject_header = h;
      break;

      /* "To:" gets flagged, and the existence of a recipient header is noted,
      whether it's resent- or not. */

    case htype_to:
      h->type = htype_to;
      /****
      to_or_cc_header_exists = TRUE;
      ****/
      break;
    }
  }

/* Extract recipients from the headers if that is required (the -t option).
Note that this is documented as being done *before* any address rewriting takes
place. There are two possibilities:

(1) According to sendmail documentation for Solaris, IRIX, and HP-UX, any
recipients already listed are to be REMOVED from the message. Smail 3 works
like this. We need to build a non-recipients tree for that list, because in
subsequent processing this data is held in a tree and that's what the
spool_write_header() function expects. Make sure that non-recipient addresses
are fully qualified and rewritten if necessary.

(2) According to other sendmail documentation, -t ADDS extracted recipients to
those in the command line arguments (and it is rumoured some other MTAs do
this). Therefore, there is an option to make Exim behave this way.

*** Notes on "Resent-" header lines ***

The presence of resent-headers in the message makes -t horribly ambiguous.
Experiments with sendmail showed that it uses recipients for all resent-
headers, totally ignoring the concept of "sets of resent- headers" as described
in RFC 2822 section 3.6.6. Sendmail also amalgamates them into a single set
with all the addresses in one instance of each header.

This seems to me not to be at all sensible. Before release 4.20, Exim 4 gave an
error for -t if there were resent- headers in the message. However, after a
discussion on the mailing list, I've learned that there are MUAs that use
resent- headers with -t, and also that the stuff about sets of resent- headers
and their ordering in RFC 2822 is generally ignored. An MUA that submits a
message with -t and resent- header lines makes sure that only *its* resent-
headers are present; previous ones are often renamed as X-resent- for example.

Consequently, Exim has been changed so that, if any resent- header lines are
present, the recipients are taken from all of the appropriate resent- lines,
and not from the ordinary To:, Cc:, etc. */

if (extract_recip)
  {
  int rcount = 0;
  error_block **bnext = &bad_addresses;

  if (extract_addresses_remove_arguments)
    {
    while (recipients_count-- > 0)
      {
      const uschar * s = rewrite_address(recipients_list[recipients_count].address,
        TRUE, TRUE, global_rewrite_rules, rewrite_existflags);
      tree_add_nonrecipient(s);
      }
    recipients_list = NULL;
    recipients_count = recipients_list_max = 0;
    }

  /* Now scan the headers */

  for (header_line * h = header_list->next; h; h = h->next)
    {
    if ((h->type == htype_to || h->type == htype_cc || h->type == htype_bcc) &&
        (!contains_resent_headers || strncmpic(h->text, US"resent-", 7) == 0))
      {
      uschar *s = Ustrchr(h->text, ':') + 1;
      while (isspace(*s)) s++;

      f.parse_allow_group = TRUE;          /* Allow address group syntax */

      while (*s != 0)
        {
        uschar *ss = parse_find_address_end(s, FALSE);
        uschar *recipient, *errmess, *pp;
        int start, end, domain;

        /* Check on maximum */

        if (recipients_max > 0 && ++rcount > recipients_max)
          give_local_error(ERRMESS_TOOMANYRECIP, US"too many recipients",
            US"message rejected: ", error_rc, stdin, NULL);
          /* Does not return */

        /* Make a copy of the address, and remove any internal newlines. These
        may be present as a result of continuations of the header line. The
        white space that follows the newline must not be removed - it is part
        of the header. */

        pp = recipient = store_get(ss - s + 1, s);
        for (uschar * p = s; p < ss; p++) if (*p != '\n') *pp++ = *p;
        *pp = 0;

#ifdef SUPPORT_I18N
        {
        BOOL b = allow_utf8_domains;
        allow_utf8_domains = TRUE;
#endif
        recipient = parse_extract_address(recipient, &errmess, &start, &end,
          &domain, FALSE);

#ifdef SUPPORT_I18N
        if (recipient)
          if (string_is_utf8(recipient)) message_smtputf8 = TRUE;
          else allow_utf8_domains = b;
        }
#else
        ;
#endif

        /* Keep a list of all the bad addresses so we can send a single
        error message at the end. However, an empty address is not an error;
        just ignore it. This can come from an empty group list like

          To: Recipients of list:;

        If there are no recipients at all, an error will occur later. */

        if (!recipient && Ustrcmp(errmess, "empty address") != 0)
          {
          int len = Ustrlen(s);
          error_block * b = store_get(sizeof(error_block), GET_UNTAINTED);
          while (len > 0 && isspace(s[len-1])) len--;
          b->next = NULL;
          b->text1 = string_printing(string_copyn(s, len));
          b->text2 = errmess;
          *bnext = b;
          bnext = &(b->next);
          }

        /* If the recipient is already in the nonrecipients tree, it must
        have appeared on the command line with the option extract_addresses_
        remove_arguments set. Do not add it to the recipients, and keep a note
        that this has happened, in order to give a better error if there are
        no recipients left. */

        else if (recipient)
          {
          if (tree_search(tree_nonrecipients, recipient) == NULL)
            receive_add_recipient(recipient, -1);
          else
            extracted_ignored = TRUE;
          }

        /* Move on past this address */

        s = ss + (*ss ? 1 : 0);
        while (isspace(*s)) s++;
        }    /* Next address */

      f.parse_allow_group = FALSE;      /* Reset group syntax flags */
      f.parse_found_group = FALSE;

      /* If this was the bcc: header, mark it "old", which means it
      will be kept on the spool, but not transmitted as part of the
      message. */

      if (h->type == htype_bcc) h->type = htype_old;
      }   /* For appropriate header line */
    }     /* For each header line */

  }

/* Now build the unique message id. This has changed several times over the
lifetime of Exim, and is changing for Exim 4.97.
The previous change was in about 2003.

Detail for the pre-4.97 version is here in [square-brackets].

The message ID has 3 parts: tttttt-ppppppppppp-ssss  (6, 11, 4 - total 23 with
the dashes).  Each part is a number in base 62.
[ tttttt-pppppp-ss  6, 6, 2 => 16 ]

The first part is the current time, in seconds.  Six chars is enough until
year 3700 with case-sensitive filesystes, but will run out in 2038 on
case-insensitive ones (Cygwin, Darwin - where we have to use base-36.
Both of those are in the "unsupported" bucket, so ignore for now).

The second part is the current pid, and supports 64b [31b] PIDs.

The third part holds sub-second time, plus (when localhost_number is set)
the host number multiplied by a number large enough to keep it away from
the time portion. Host numbers are restricted to the range 0-16.
The time resolution is variously 1, 2 or 4 microseconds [0.5 or 1 ms]
depending on the use of localhost_nubmer and of case-insensitive filesystems.

After a message has been received, Exim ensures that the timer has ticked at the
appropriate level before proceeding, to avoid duplication if the pid happened to
be re-used within the same time period. It seems likely that most messages will
take at least half a millisecond to be received, so no delay will normally be
necessary. At least for some time...

Note that string_base62_XX() returns its data in a static storage block, so it
must be copied before calling string_base62_XXX) again. It always returns exactly
11 (_64) or 6 (_32) characters.

There doesn't seem to be anything in the RFC which requires a message id to
start with a letter, but Smail was changed to ensure this. The external form of
the message id (as supplied by string expansion) therefore starts with an
additional leading 'E'. The spool file names do not include this leading
letter and it is not used internally.

NOTE: If ever the format of message ids is changed, the regular expression for
checking that a string is in this format must be updated in a corresponding
way. It appears in the initializing code in exim.c. The macro MESSAGE_ID_LENGTH
must also be changed to reflect the correct string length. The queue-sort code
needs to know the layout. Then, of course, other programs that rely on the
message id format will need updating too (inc. at least exim_msgdate). */

Ustrncpy(message_id, string_base62_32((long int)(message_id_tv.tv_sec)), MESSAGE_ID_TIME_LEN);
message_id[MESSAGE_ID_TIME_LEN] = '-';
Ustrncpy(message_id + MESSAGE_ID_TIME_LEN + 1,
        string_base62_64((long int)getpid()),
        MESSAGE_ID_PID_LEN
        );

/* Deal with the case where the host number is set. The value of the number was
checked when it was read, to ensure it isn't too big. */

if (host_number_string)
  sprintf(CS(message_id + MESSAGE_ID_TIME_LEN + 1 + MESSAGE_ID_PID_LEN),
        "-%" str(MESSAGE_ID_SUBTIME_LEN) "s",
        string_base62_32((long int)(
          host_number * (1000000/id_resolution)
          + message_id_tv.tv_usec/id_resolution))
        + (6 - MESSAGE_ID_SUBTIME_LEN)
         );

/* Host number not set: final field is just the fractional time at an
appropriate resolution. */

else
  sprintf(CS(message_id + MESSAGE_ID_TIME_LEN + 1 + MESSAGE_ID_PID_LEN),
    "-%" str(MESSAGE_ID_SUBTIME_LEN) "s",
        string_base62_32((long int)(message_id_tv.tv_usec/id_resolution))
        + (6 - MESSAGE_ID_SUBTIME_LEN));

/* Add the current message id onto the current process info string if
it will fit. */

(void)string_format(process_info + process_info_len,
  PROCESS_INFO_SIZE - process_info_len, " id=%s", message_id);

/* If we are using multiple input directories, set up the one for this message
to be the least significant base-62 digit of the time of arrival. Otherwise
ensure that it is an empty string. */

set_subdir_str(message_subdir, message_id, 0);

/* Now that we have the message-id, if there is no message-id: header, generate
one, but only for local (without suppress_local_fixups) or submission mode
messages. This can be user-configured if required, but we had better flatten
any illegal characters therein. */

if (  !msgid_header
   && ((!sender_host_address && !f.suppress_local_fixups) || f.submission_mode))
  {
  uschar *id_text = US"";
  uschar *id_domain = primary_hostname;
  header_line * h;

  /* Permit only letters, digits, dots, and hyphens in the domain */

  if (message_id_domain)
    {
    uschar *new_id_domain = expand_string(message_id_domain);
    if (!new_id_domain)
      {
      if (!f.expand_string_forcedfail)
        log_write(0, LOG_MAIN|LOG_PANIC,
          "expansion of \"%s\" (message_id_header_domain) "
          "failed: %s", message_id_domain, expand_string_message);
      }
    else if (*new_id_domain)
      {
      id_domain = new_id_domain;
      for (uschar * p = id_domain; *p; p++)
        if (!isalnum(*p) && *p != '.') *p = '-';  /* No need to test '-' ! */
      }
    }

  /* Permit all characters except controls and RFC 2822 specials in the
  additional text part. */

  if (message_id_text)
    {
    uschar *new_id_text = expand_string(message_id_text);
    if (!new_id_text)
      {
      if (!f.expand_string_forcedfail)
        log_write(0, LOG_MAIN|LOG_PANIC,
          "expansion of \"%s\" (message_id_header_text) "
          "failed: %s", message_id_text, expand_string_message);
      }
    else if (*new_id_text)
      {
      id_text = new_id_text;
      for (uschar * p = id_text; *p; p++) if (mac_iscntrl_or_special(*p)) *p = '-';
      }
    }

  /* Add the header line.
  Resent-* headers are prepended, per RFC 5322 3.6.6.  Non-Resent-* are
  appended, to preserve classical expectations of header ordering. */

  h = header_add_at_position_internal(!resents_exist, NULL, FALSE, htype_id,
    "%sMessage-Id: <%s%s%s@%s>\n", resent_prefix, message_id_external,
    *id_text == 0 ? "" : ".", id_text, id_domain);

  /* Arrange for newly-created Message-Id to be logged */

  if (!resents_exist)
    {
    msgid_header_newly_created = TRUE;
    msgid_header = h;
    }
  }

/* If we are to log recipients, keep a copy of the raw ones before any possible
rewriting. Must copy the count, because later ACLs and the local_scan()
function may mess with the real recipients. */

if (LOGGING(received_recipients))
  {
  raw_recipients = store_get(recipients_count * sizeof(uschar *), GET_UNTAINTED);
  for (int i = 0; i < recipients_count; i++)
    raw_recipients[i] = string_copy(recipients_list[i].address);
  raw_recipients_count = recipients_count;
  }

/* Ensure the recipients list is fully qualified and rewritten. Unqualified
recipients will get here only if the conditions were right (allow_unqualified_
recipient is TRUE). */

DEBUG(D_rewrite)
  { debug_printf_indent("qualify & rewrite recipients list\n"); acl_level++; }
for (int i = 0; i < recipients_count; i++)
  recipients_list[i].address =  /* deconst ok as src was not cont */
    US rewrite_address(recipients_list[i].address, TRUE, TRUE,
      global_rewrite_rules, rewrite_existflags);
DEBUG(D_rewrite) acl_level--;

/* If there is no From: header, generate one for local (without
suppress_local_fixups) or submission_mode messages. If there is no sender
address, but the sender is local or this is a local delivery error, use the
originator login. This shouldn't happen for genuine bounces, but might happen
for autoreplies. The addition of From: must be done *before* checking for the
possible addition of a Sender: header, because untrusted_set_sender allows an
untrusted user to set anything in the envelope (which might then get info
From:) but we still want to ensure a valid Sender: if it is required. */

if (  !from_header
   && ((!sender_host_address && !f.suppress_local_fixups) || f.submission_mode))
  {
  const uschar * oname = US"";

  /* Use the originator_name if this is a locally submitted message and the
  caller is not trusted. For trusted callers, use it only if -F was used to
  force its value or if we have a non-SMTP message for which -f was not used
  to set the sender. */

  if (!sender_host_address)
    {
    if (!f.trusted_caller || f.sender_name_forced ||
         (!smtp_input && !f.sender_address_forced))
      oname = originator_name;
    }

  /* For non-locally submitted messages, the only time we use the originator
  name is when it was forced by the /name= option on control=submission. */

  else if (submission_name) oname = submission_name;

  /* Envelope sender is empty */

  if (!*sender_address)
    {
    uschar *fromstart, *fromend;

    fromstart = string_sprintf("%sFrom: %s%s",
      resent_prefix, oname, *oname ? " <" : "");
    fromend = *oname ? US">" : US"";

    if (f.sender_local || f.local_error_message)
      header_add(htype_from, "%s%s@%s%s\n", fromstart,
        local_part_quote(originator_login), qualify_domain_sender,
        fromend);

    else if (f.submission_mode && authenticated_id)
      {
      if (!submission_domain)
        header_add(htype_from, "%s%s@%s%s\n", fromstart,
          local_part_quote(authenticated_id), qualify_domain_sender,
          fromend);

      else if (!*submission_domain)  /* empty => whole address set */
        header_add(htype_from, "%s%s%s\n", fromstart, authenticated_id,
          fromend);

      else
        header_add(htype_from, "%s%s@%s%s\n", fromstart,
          local_part_quote(authenticated_id), submission_domain, fromend);

      from_header = header_last;    /* To get it checked for Sender: */
      }
    }

  /* There is a non-null envelope sender. Build the header using the original
  sender address, before any rewriting that might have been done while
  verifying it. */

  else
    {
    header_add(htype_from, "%sFrom: %s%s%s%s\n", resent_prefix,
      oname,
      *oname ? " <" : "",
      sender_address_unrewritten ? sender_address_unrewritten : sender_address,
      *oname ? ">" : "");

    from_header = header_last;    /* To get it checked for Sender: */
    }
  }


/* If the sender is local (without suppress_local_fixups), or if we are in
submission mode and there is an authenticated_id, check that an existing From:
is correct, and if not, generate a Sender: header, unless disabled. Any
previously-existing Sender: header was removed above. Note that sender_local,
as well as being TRUE if the caller of exim is not trusted, is also true if a
trusted caller did not supply a -f argument for non-smtp input. To allow
trusted callers to forge From: without supplying -f, we have to test explicitly
here. If the From: header contains more than one address, then the call to
parse_extract_address fails, and a Sender: header is inserted, as required. */

if (  from_header
   && (  f.active_local_from_check
      && (  f.sender_local && !f.trusted_caller && !f.suppress_local_fixups
         || f.submission_mode && authenticated_id
   )  )  )
  {
  BOOL make_sender = TRUE;
  int start, end, domain;
  uschar *errmess;
  uschar *from_address =
    parse_extract_address(Ustrchr(from_header->text, ':') + 1, &errmess,
      &start, &end, &domain, FALSE);
  uschar *generated_sender_address;

  generated_sender_address = f.submission_mode
    ? !submission_domain
    ? string_sprintf("%s@%s",
        local_part_quote(authenticated_id), qualify_domain_sender)
    : !*submission_domain                       /* empty => full address */
    ? string_sprintf("%s", authenticated_id)
    : string_sprintf("%s@%s",
        local_part_quote(authenticated_id), submission_domain)
    : string_sprintf("%s@%s",
        local_part_quote(originator_login), qualify_domain_sender);

  /* Remove permitted prefixes and suffixes from the local part of the From:
  address before doing the comparison with the generated sender. */

  if (from_address)
    {
    int slen;
    uschar *at = domain ? from_address + domain - 1 : NULL;

    if (at) *at = 0;
    from_address += route_check_prefix(from_address, local_from_prefix, NULL);
    if ((slen = route_check_suffix(from_address, local_from_suffix, NULL)) > 0)
      {
      memmove(from_address+slen, from_address, Ustrlen(from_address)-slen);
      from_address += slen;
      }
    if (at) *at = '@';

    if (  strcmpic(generated_sender_address, from_address) == 0
       || (!domain && strcmpic(from_address, originator_login) == 0))
        make_sender = FALSE;
    }

  /* We have to cause the Sender header to be rewritten if there are
  appropriate rewriting rules. */

  if (make_sender)
    if (f.submission_mode && !submission_name)
      header_add(htype_sender, "%sSender: %s\n", resent_prefix,
        generated_sender_address);
    else
      header_add(htype_sender, "%sSender: %s <%s>\n",
        resent_prefix,
        f.submission_mode ? submission_name : originator_name,
        generated_sender_address);

  /* Ensure that a non-null envelope sender address corresponds to the
  submission mode sender address. */

  if (f.submission_mode && *sender_address)
    {
    if (!sender_address_unrewritten)
      sender_address_unrewritten = sender_address;
    sender_address = generated_sender_address;
    if (Ustrcmp(sender_address_unrewritten, generated_sender_address) != 0)
      log_write(L_address_rewrite, LOG_MAIN,
        "\"%s\" from env-from rewritten as \"%s\" by submission mode",
        sender_address_unrewritten, generated_sender_address);
    }
  }

/* If there are any rewriting rules, apply them to the sender address, unless
it has already been rewritten as part of verification for SMTP input. */

DEBUG(D_rewrite)
  { debug_printf("global rewrite rules\n"); acl_level++; }
if (global_rewrite_rules && !sender_address_unrewritten && *sender_address)
  {
  /* deconst ok as src was not const */
  sender_address = US rewrite_address(sender_address, FALSE, TRUE,
    global_rewrite_rules, rewrite_existflags);
  DEBUG(D_receive|D_rewrite)
    debug_printf("rewritten sender = %s\n", sender_address);
  }
DEBUG(D_rewrite) acl_level--;


/* The headers must be run through rewrite_header(), because it ensures that
addresses are fully qualified, as well as applying any rewriting rules that may
exist.

Qualification of header addresses in a message from a remote host happens only
if the host is in sender_unqualified_hosts or recipient_unqualified hosts, as
appropriate. For local messages, qualification always happens, unless -bnq is
used to explicitly suppress it. No rewriting is done for an unqualified address
that is left untouched.

We start at the second header, skipping our own Received:. This rewriting is
documented as happening *after* recipient addresses are taken from the headers
by the -t command line option. An added Sender: gets rewritten here. */

DEBUG(D_rewrite)
  { debug_printf("rewrite headers\n"); acl_level++; }
for (header_line * h = header_list->next, * newh; h; h = h->next)
  if ((newh = rewrite_header(h, NULL, NULL, global_rewrite_rules,
                              rewrite_existflags, TRUE)))
    h = newh;
DEBUG(D_rewrite) acl_level--;


/* An RFC 822 (sic) message is not legal unless it has at least one of "to",
"cc", or "bcc". Note that although the minimal examples in RFC 822 show just
"to" or "bcc", the full syntax spec allows "cc" as well. If any resent- header
exists, this applies to the set of resent- headers rather than the normal set.

The requirement for a recipient header has been removed in RFC 2822. At this
point in the code, earlier versions of Exim added a To: header for locally
submitted messages, and an empty Bcc: header for others. In the light of the
changes in RFC 2822, this was dropped in November 2003. */


/* If there is no date header, generate one if the message originates locally
(i.e. not over TCP/IP) and suppress_local_fixups is not set, or if the
submission mode flag is set. Messages without Date: are not valid, but it seems
to be more confusing if Exim adds one to all remotely-originated messages.
As per Message-Id, we prepend if resending, else append.
*/

if (  !date_header_exists
   && ((!sender_host_address && !f.suppress_local_fixups) || f.submission_mode))
  header_add_at_position(!resents_exist, NULL, FALSE, htype_other,
    "%sDate: %s\n", resent_prefix, tod_stamp(tod_full));

search_tidyup();    /* Free any cached resources */

/* Show the complete set of headers if debugging. Note that the first one (the
new Received:) has not yet been set. */

DEBUG(D_receive)
  {
  debug_printf(">>Headers after rewriting and local additions:\n");
  for (header_line * h = header_list->next; h; h = h->next)
    debug_printf("%c %s", h->type, h->text);
  debug_printf("\n");
  }

/* The headers are now complete in store. If we are running in filter
testing mode, that is all this function does. Return TRUE if the message
ended with a dot. */

if (filter_test != FTEST_NONE)
  {
  process_info[process_info_len] = 0;
  return message_ended == END_DOT;
  }

/*XXX CHUNKING: need to cancel cutthrough under BDAT, for now.  In future,
think more if it could be handled.  Cannot do onward CHUNKING unless
inbound is, but inbound chunking ought to be ok with outbound plain.
Could we do onward CHUNKING given inbound CHUNKING?
*/
if (chunking_state > CHUNKING_OFFERED)
  cancel_cutthrough_connection(FALSE, US"chunking active");

/* Cutthrough delivery:
We have to create the Received header now rather than at the end of reception,
so the timestamp behaviour is a change to the normal case.
Having created it, send the headers to the destination. */

if (cutthrough.cctx.sock >= 0 && cutthrough.delivery)
  {
  if (received_count > received_headers_max)
    {
    cancel_cutthrough_connection(TRUE, US"too many headers");
    if (smtp_input) receive_swallow_smtp();  /* Swallow incoming SMTP */
    log_write(0, LOG_MAIN|LOG_REJECT, "rejected from <%s>%s%s%s%s: "
      "Too many \"Received\" headers",
      sender_address,
      sender_fullhost ? "H=" : "", sender_fullhost ? sender_fullhost : US"",
      sender_ident ? "U=" : "", sender_ident ? sender_ident : US"");
    smtp_reply = US"550 Too many \"Received\" headers - suspected mail loop";
    goto NOT_ACCEPTED;                          /* Skip to end of function */
    }
  received_header_gen();
  add_acl_headers(ACL_WHERE_RCPT, US"MAIL or RCPT");
  (void) cutthrough_headers_send();
  }


/* Open a new spool file for the data portion of the message. We need
to access it both via a file descriptor and a stream. Try to make the
directory if it isn't there. */

spool_name = spool_fname(US"input", message_subdir, message_id, US"-D");
DEBUG(D_receive) debug_printf("Data file name: %s\n", spool_name);

if ((data_fd = Uopen(spool_name, O_RDWR|O_CREAT|O_EXCL, SPOOL_MODE)) < 0)
  {
  if (errno == ENOENT)
    {
    (void) directory_make(spool_directory,
                        spool_sname(US"input", message_subdir),
                        INPUT_DIRECTORY_MODE, TRUE);
    data_fd = Uopen(spool_name, O_RDWR|O_CREAT|O_EXCL, SPOOL_MODE);
    }
  if (data_fd < 0)
    log_write(0, LOG_MAIN|LOG_PANIC_DIE, "Failed to create spool file %s: %s",
      spool_name, strerror(errno));
  }

/* Make sure the file's group is the Exim gid, and double-check the mode
because the group setting doesn't always get set automatically. */

if (0 != exim_fchown(data_fd, exim_uid, exim_gid, spool_name))
  log_write(0, LOG_MAIN|LOG_PANIC_DIE,
    "Failed setting ownership on spool file %s: %s",
    spool_name, strerror(errno));
(void)fchmod(data_fd, SPOOL_MODE);

/* We now have data file open. Build a stream for it and lock it. We lock only
the first line of the file (containing the message ID) because otherwise there
are problems when Exim is run under Cygwin (I'm told). See comments in
spool_in.c, where the same locking is done. */

spool_data_file = fdopen(data_fd, "w+");
lock_data.l_type = F_WRLCK;
lock_data.l_whence = SEEK_SET;
lock_data.l_start = 0;
lock_data.l_len = spool_data_start_offset(message_id);

if (fcntl(data_fd, F_SETLK, &lock_data) < 0)
  log_write(0, LOG_MAIN|LOG_PANIC_DIE, "Cannot lock %s (%d): %s", spool_name,
    errno, strerror(errno));

/* We have an open, locked data file. Write the message id to it to make it
self-identifying. Then read the remainder of the input of this message and
write it to the data file. If the variable next != NULL, it contains the first
data line (which was read as a header but then turned out not to have the right
format); write it (remembering that it might contain binary zeros). The result
of fwrite() isn't inspected; instead we call ferror() below. */

fprintf(spool_data_file, "%s-D\n", message_id);
if (next)
  {
  uschar *s = next->text;
  int len = next->slen;
  if (fwrite(s, 1, len, spool_data_file) == len) /* "if" for compiler quietening */
    body_linecount++;                 /* Assumes only 1 line */
  }

/* Note that we might already be at end of file, or the logical end of file
(indicated by '.'), or might have encountered an error while writing the
message id or "next" line. */

if (!ferror(spool_data_file) && !(receive_feof)() && message_ended != END_DOT)
  {
  if (smtp_input)
    {
    message_ended = chunking_state <= CHUNKING_OFFERED
      ? read_message_data_smtp(spool_data_file)
      : spool_wireformat
      ? read_message_bdat_smtp_wire(spool_data_file)
      : read_message_bdat_smtp(spool_data_file);
    receive_linecount++;                /* The terminating "." line */
    }
  else
    message_ended = read_message_data(spool_data_file);

  receive_linecount += body_linecount;  /* For BSMTP errors mainly */
  message_linecount += body_linecount;

  switch (message_ended)
    {
    /* Handle premature termination of SMTP */

    case END_EOF:
      if (smtp_input)
        {
        Uunlink(spool_name);            /* Lose data file when closed */
        cancel_cutthrough_connection(TRUE, US"sender closed connection");
        smtp_reply = handle_lost_connection(US"");
        smtp_yield = FALSE;
        goto NOT_ACCEPTED;                              /* Skip to end of function */
        }
      break;

    /* Handle message that is too big. Don't use host_or_ident() in the log
    message; we want to see the ident value even for non-remote messages. */

    case END_SIZE:
      Uunlink(spool_name);                /* Lose the data file when closed */
      cancel_cutthrough_connection(TRUE, US"mail too big");
      if (smtp_input) receive_swallow_smtp();  /* Swallow incoming SMTP */

      log_write(L_size_reject, LOG_MAIN|LOG_REJECT, "rejected from <%s>%s%s%s%s: "
        "message too big: read=%d max=%d",
        sender_address,
        sender_fullhost ? " H=" : "",
        sender_fullhost ? sender_fullhost : US"",
        sender_ident ? " U=" : "",
        sender_ident ? sender_ident : US"",
        message_size,
        thismessage_size_limit);

      if (smtp_input)
        {
        smtp_reply = US"552 Message size exceeds maximum permitted";
        goto NOT_ACCEPTED;                      /* Skip to end of function */
        }
      else
        {
        fseek(spool_data_file, (long int)spool_data_start_offset(message_id), SEEK_SET);
        give_local_error(ERRMESS_TOOBIG,
          string_sprintf("message too big (max=%d)", thismessage_size_limit),
          US"message rejected: ", error_rc, spool_data_file, header_list);
        /* Does not return */
        }
      break;

    /* Handle bad BDAT protocol sequence */

    case END_PROTOCOL:
      Uunlink(spool_name);              /* Lose the data file when closed */
      cancel_cutthrough_connection(TRUE, US"sender protocol error");
      smtp_reply = US"";                /* Response already sent */
      goto NOT_ACCEPTED;                        /* Skip to end of function */
    }
  }

/* Restore the standard SIGALRM handler for any subsequent processing. (For
example, there may be some expansion in an ACL that uses a timer.) */

os_non_restarting_signal(SIGALRM, sigalrm_handler);

/* The message body has now been read into the data file. Call fflush() to
empty the buffers in C, and then call fsync() to get the data written out onto
the disk, as fflush() doesn't do this (or at least, it isn't documented as
having to do this). If there was an I/O error on either input or output,
attempt to send an error message, and unlink the spool file. For non-SMTP input
we can then give up. Note that for SMTP input we must swallow the remainder of
the input in cases of output errors, since the far end doesn't expect to see
anything until the terminating dot line is sent. */

if (fflush(spool_data_file) == EOF || ferror(spool_data_file) ||
    EXIMfsync(fileno(spool_data_file)) < 0 || (receive_ferror)())
  {
  uschar *msg_errno = US strerror(errno);
  BOOL input_error = (receive_ferror)() != 0;
  uschar *msg = string_sprintf("%s error (%s) while receiving message from %s",
    input_error? "Input read" : "Spool write",
    msg_errno,
    sender_fullhost ? sender_fullhost : sender_ident);

  log_write(0, LOG_MAIN, "Message abandoned: %s", msg);
  Uunlink(spool_name);                /* Lose the data file */
  cancel_cutthrough_connection(TRUE, US"error writing spoolfile");

  if (smtp_input)
    {
    if (input_error)
      smtp_reply = US"451 Error while reading input data";
    else
      {
      smtp_reply = US"451 Error while writing spool file";
      receive_swallow_smtp();
      }
    goto NOT_ACCEPTED;                  /* Skip to end of function */
    }

  else
    {
    fseek(spool_data_file, (long int)spool_data_start_offset(message_id), SEEK_SET);
    give_local_error(ERRMESS_IOERR, msg, US"", error_rc, spool_data_file,
      header_list);
    /* Does not return */
    }
  }


/* No I/O errors were encountered while writing the data file. */

DEBUG(D_receive) debug_printf("Data file written for message %s\n", message_id);
gettimeofday(&received_time_complete, NULL);


/* If there were any bad addresses extracted by -t, or there were no recipients
left after -t, send a message to the sender of this message, or write it to
stderr if the error handling option is set that way. Note that there may
legitimately be no recipients for an SMTP message if they have all been removed
by "discard".

We need to rewind the data file in order to read it. In the case of no
recipients or stderr error writing, throw the data file away afterwards, and
exit. (This can't be SMTP, which always ensures there's at least one
syntactically good recipient address.) */

if (extract_recip && (bad_addresses || recipients_count == 0))
  {
  DEBUG(D_receive)
    {
    if (recipients_count == 0) debug_printf("*** No recipients\n");
    if (bad_addresses)
      {
      debug_printf("*** Bad address(es)\n");
      for (error_block * eblock = bad_addresses; eblock; eblock = eblock->next)
        debug_printf("  %s: %s\n", eblock->text1, eblock->text2);
      }
    }

  log_write(0, LOG_MAIN|LOG_PANIC, "%s found in headers",
    bad_addresses ? "bad addresses" : "no recipients");

  fseek(spool_data_file, (long int)spool_data_start_offset(message_id), SEEK_SET);

  /* If configured to send errors to the sender, but this fails, force
  a failure error code. We use a special one for no recipients so that it
  can be detected by the autoreply transport. Otherwise error_rc is set to
  errors_sender_rc, which is EXIT_FAILURE unless -oee was given, in which case
  it is EXIT_SUCCESS. */

  if (error_handling == ERRORS_SENDER)
    {
    if (!moan_to_sender(
          bad_addresses
          ? recipients_list ? ERRMESS_BADADDRESS : ERRMESS_BADNOADDRESS
          : extracted_ignored ? ERRMESS_IGADDRESS : ERRMESS_NOADDRESS,
          bad_addresses, header_list, spool_data_file, FALSE
       )               )
      error_rc = bad_addresses ? EXIT_FAILURE : EXIT_NORECIPIENTS;
    }
  else
    {
    if (!bad_addresses)
      if (extracted_ignored)
        fprintf(stderr, "exim: all -t recipients overridden by command line\n");
      else
        fprintf(stderr, "exim: no recipients in message\n");
    else
      {
      fprintf(stderr, "exim: invalid address%s",
        bad_addresses->next ? "es:\n" : ":");
      for ( ; bad_addresses; bad_addresses = bad_addresses->next)
        fprintf(stderr, "  %s: %s\n", bad_addresses->text1,
          bad_addresses->text2);
      }
    }

  if (recipients_count == 0 || error_handling == ERRORS_STDERR)
    {
    Uunlink(spool_name);
    (void)fclose(spool_data_file);
    exim_exit(error_rc);
    }
  }

/* Data file successfully written. Generate text for the Received: header by
expanding the configured string, and adding a timestamp. By leaving this
operation till now, we ensure that the timestamp is the time that message
reception was completed. However, this is deliberately done before calling the
data ACL and local_scan().

This Received: header may therefore be inspected by the data ACL and by code in
the local_scan() function. When they have run, we update the timestamp to be
the final time of reception.

If there is just one recipient, set up its value in the $received_for variable
for use when we generate the Received: header.

Note: the checking for too many Received: headers is handled by the delivery
code. */
/*XXX eventually add excess Received: check for cutthrough case back when classifying them */

if (!received_header->text)     /* Non-cutthrough case */
  {
  received_header_gen();

  /* Set the value of message_body_size for the DATA ACL and for local_scan() */

  message_body_size = (fstat(data_fd, &statbuf) == 0)?
    statbuf.st_size - spool_data_start_offset(message_id) : -1;

  /* If an ACL from any RCPT commands set up any warning headers to add, do so
  now, before running the DATA ACL. */

  add_acl_headers(ACL_WHERE_RCPT, US"MAIL or RCPT");
  }
else
  message_body_size = (fstat(data_fd, &statbuf) == 0)?
    statbuf.st_size - spool_data_start_offset(message_id) : -1;

/* If an ACL is specified for checking things at this stage of reception of a
message, run it, unless all the recipients were removed by "discard" in earlier
ACLs. That is the only case in which recipients_count can be zero at this
stage. Set deliver_datafile to point to the data file so that $message_body and
$message_body_end can be extracted if needed. Allow $recipients in expansions.
*/

deliver_datafile = data_fd;
user_msg = NULL;

f.enable_dollar_recipients = TRUE;

if (recipients_count == 0)
  blackholed_by = f.recipients_discarded ? US"MAIL ACL" : US"RCPT ACL";

else
  {
  /* Handle interactive SMTP messages */

  if (smtp_input && !smtp_batched_input)
    {

#ifndef DISABLE_DKIM
    if (!f.dkim_disable_verify)
      {
      /* Finish verification */
      dkim_exim_verify_finish();

      /* Check if we must run the DKIM ACL */
      if (acl_smtp_dkim && dkim_verify_signers && *dkim_verify_signers)
        {
        uschar * dkim_verify_signers_expanded =
          expand_string(dkim_verify_signers);
        gstring * results = NULL;
        int signer_sep = 0;
        const uschar * ptr;
        uschar * item;
        gstring * seen_items = NULL;
        int old_pool = store_pool;

        store_pool = POOL_PERM;   /* Allow created variables to live to data ACL */

        if (!(ptr = dkim_verify_signers_expanded))
          log_write(0, LOG_MAIN|LOG_PANIC,
            "expansion of dkim_verify_signers option failed: %s",
            expand_string_message);

        /* Default to OK when no items are present */
        rc = OK;
        while ((item = string_nextinlist(&ptr, &signer_sep, NULL, 0)))
          {
          /* Prevent running ACL for an empty item */
          if (!item || !*item) continue;

          /* Only run ACL once for each domain or identity,
          no matter how often it appears in the expanded list. */
          if (seen_items)
            {
            uschar * seen_item;
            const uschar * seen_items_list = string_from_gstring(seen_items);
            int seen_sep = ':';
            BOOL seen_this_item = FALSE;

            while ((seen_item = string_nextinlist(&seen_items_list, &seen_sep,
                                                  NULL, 0)))
              if (Ustrcmp(seen_item,item) == 0)
                {
                seen_this_item = TRUE;
                break;
                }

            if (seen_this_item)
              {
              DEBUG(D_receive)
                debug_printf("acl_smtp_dkim: skipping signer %s, "
                  "already seen\n", item);
              continue;
              }

            seen_items = string_catn(seen_items, US":", 1);
            }
          seen_items = string_cat(seen_items, item);

          rc = dkim_exim_acl_run(item, &results, &user_msg, &log_msg);
          if (rc != OK)
            {
            DEBUG(D_receive)
              debug_printf("acl_smtp_dkim: acl_check returned %d on %s, "
                "skipping remaining items\n", rc, item);
            cancel_cutthrough_connection(TRUE, US"dkim acl not ok");
            break;
            }
          }
        dkim_verify_status = string_from_gstring(results);
        store_pool = old_pool;
        add_acl_headers(ACL_WHERE_DKIM, US"DKIM");
        if (rc == DISCARD)
          {
          recipients_count = 0;
          blackholed_by = US"DKIM ACL";
          if (log_msg)
            blackhole_log_msg = string_sprintf(": %s", log_msg);
          }
        else if (rc != OK)
          {
          Uunlink(spool_name);
          if (smtp_handle_acl_fail(ACL_WHERE_DKIM, rc, user_msg, log_msg) != 0)
            smtp_yield = FALSE;    /* No more messages after dropped connection */
          smtp_reply = US"";       /* Indicate reply already sent */
          goto NOT_ACCEPTED;                    /* Skip to end of function */
          }
        }
      else
        dkim_exim_verify_log_all();
      }
#endif /* DISABLE_DKIM */

#ifdef WITH_CONTENT_SCAN
    if (  recipients_count > 0
       && acl_smtp_mime
       && !run_mime_acl(acl_smtp_mime, &smtp_yield, &smtp_reply, &blackholed_by)
       )
      goto TIDYUP;
#endif /* WITH_CONTENT_SCAN */

#ifdef SUPPORT_DMARC
    dmarc_store_data(from_header);
#endif

#ifndef DISABLE_PRDR
    if (prdr_requested && recipients_count > 1 && acl_smtp_data_prdr)
      {
      int all_pass = OK;
      int all_fail = FAIL;

      smtp_printf("353 PRDR content analysis beginning\r\n", SP_MORE);
      /* Loop through recipients, responses must be in same order received */
      for (unsigned int c = 0; recipients_count > c; c++)
        {
        const uschar * addr = recipients_list[c].address;
        uschar * msg= US"PRDR R=<%s> %s";
        uschar * code;
        DEBUG(D_receive)
          debug_printf("PRDR processing recipient %s (%d of %d)\n",
                       addr, c+1, recipients_count);
        rc = acl_check(ACL_WHERE_PRDR, addr,
                       acl_smtp_data_prdr, &user_msg, &log_msg);

        /* If any recipient rejected content, indicate it in final message */
        all_pass |= rc;
        /* If all recipients rejected, indicate in final message */
        all_fail &= rc;

        switch (rc)
          {
          case OK: case DISCARD: code = US"250"; break;
          case DEFER:            code = US"450"; break;
          default:               code = US"550"; break;
          }
        if (user_msg != NULL)
          smtp_user_msg(code, user_msg);
        else
          {
          switch (rc)
            {
            case OK: case DISCARD:
              msg = string_sprintf(CS msg, addr, "acceptance");        break;
            case DEFER:
              msg = string_sprintf(CS msg, addr, "temporary refusal"); break;
            default:
              msg = string_sprintf(CS msg, addr, "refusal");           break;
            }
          smtp_user_msg(code, msg);
          }
        if (log_msg)       log_write(0, LOG_MAIN, "PRDR %s %s", addr, log_msg);
        else if (user_msg) log_write(0, LOG_MAIN, "PRDR %s %s", addr, user_msg);
        else               log_write(0, LOG_MAIN, "%s", CS msg);

        if (rc != OK) { receive_remove_recipient(addr); c--; }
        }
      /* Set up final message, used if data acl gives OK */
      smtp_reply = string_sprintf("%s id=%s message %s",
                       all_fail == FAIL ? US"550" : US"250",
                       message_id,
                       all_fail == FAIL
                         ? US"rejected for all recipients"
                         : all_pass == OK
                           ? US"accepted"
                           : US"accepted for some recipients");
      if (recipients_count == 0)
        goto NOT_ACCEPTED;
      }
    else
      prdr_requested = FALSE;
#endif /* !DISABLE_PRDR */

    /* Check the recipients count again, as the MIME ACL might have changed
    them. */

    if (acl_smtp_data && recipients_count > 0)
      {
      rc = acl_check(ACL_WHERE_DATA, NULL, acl_smtp_data, &user_msg, &log_msg);
      add_acl_headers(ACL_WHERE_DATA, US"DATA");
      if (rc == DISCARD)
        {
        recipients_count = 0;
        blackholed_by = US"DATA ACL";
        if (log_msg)
          blackhole_log_msg = string_sprintf(": %s", log_msg);
        cancel_cutthrough_connection(TRUE, US"data acl discard");
        }
      else if (rc != OK)
        {
        Uunlink(spool_name);
        cancel_cutthrough_connection(TRUE, US"data acl not ok");
#ifdef WITH_CONTENT_SCAN
        unspool_mbox();
#endif
#ifdef EXPERIMENTAL_DCC
        dcc_ok = 0;
#endif
        if (smtp_handle_acl_fail(ACL_WHERE_DATA, rc, user_msg, log_msg) != 0)
          smtp_yield = FALSE;    /* No more messages after dropped connection */
        smtp_reply = US"";       /* Indicate reply already sent */
        goto NOT_ACCEPTED;                      /* Skip to end of function */
        }
      }
    }

  /* Handle non-SMTP and batch SMTP (i.e. non-interactive) messages. Note that
  we cannot take different actions for permanent and temporary rejections. */

  else
    {

#ifdef WITH_CONTENT_SCAN
    if (  acl_not_smtp_mime
       && !run_mime_acl(acl_not_smtp_mime, &smtp_yield, &smtp_reply,
          &blackholed_by)
       )
      goto TIDYUP;
#endif /* WITH_CONTENT_SCAN */

    if (acl_not_smtp)
      {
      uschar *user_msg, *log_msg;
      f.authentication_local = TRUE;
      rc = acl_check(ACL_WHERE_NOTSMTP, NULL, acl_not_smtp, &user_msg, &log_msg);
      if (rc == DISCARD)
        {
        recipients_count = 0;
        blackholed_by = US"non-SMTP ACL";
        if (log_msg)
          blackhole_log_msg = string_sprintf(": %s", log_msg);
        }
      else if (rc != OK)
        {
        Uunlink(spool_name);
#ifdef WITH_CONTENT_SCAN
        unspool_mbox();
#endif
#ifdef EXPERIMENTAL_DCC
        dcc_ok = 0;
#endif
        /* The ACL can specify where rejections are to be logged, possibly
        nowhere. The default is main and reject logs. */

        if (log_reject_target)
          log_write(0, log_reject_target, "F=<%s> rejected by non-SMTP ACL: %s",
            sender_address, log_msg);

        if (!user_msg) user_msg = US"local configuration problem";
        if (smtp_batched_input)
          moan_smtp_batch(NULL, "%d %s", 550, user_msg);
          /* Does not return */
        else
          {
          fseek(spool_data_file, (long int)spool_data_start_offset(message_id), SEEK_SET);
          give_local_error(ERRMESS_LOCAL_ACL, user_msg,
            US"message rejected by non-SMTP ACL: ", error_rc, spool_data_file,
              header_list);
          /* Does not return */
          }
        }
      add_acl_headers(ACL_WHERE_NOTSMTP, US"non-SMTP");
      }
    }

  /* The applicable ACLs have been run */

  if (f.deliver_freeze) frozen_by = US"ACL";     /* for later logging */
  if (f.queue_only_policy) queued_by = US"ACL";
  }

#ifdef WITH_CONTENT_SCAN
unspool_mbox();
#endif

#ifdef EXPERIMENTAL_DCC
dcc_ok = 0;
#endif


#ifdef HAVE_LOCAL_SCAN
/* The final check on the message is to run the scan_local() function. The
version supplied with Exim always accepts, but this is a hook for sysadmins to
supply their own checking code. The local_scan() function is run even when all
the recipients have been discarded. */

lseek(data_fd, (long int)spool_data_start_offset(message_id), SEEK_SET);

/* Arrange to catch crashes in local_scan(), so that the -D file gets
deleted, and the incident gets logged. */

if (sigsetjmp(local_scan_env, 1) == 0)
  {
  had_local_scan_crash = 0;
  os_non_restarting_signal(SIGSEGV, local_scan_crash_handler);
  os_non_restarting_signal(SIGFPE, local_scan_crash_handler);
  os_non_restarting_signal(SIGILL, local_scan_crash_handler);
  os_non_restarting_signal(SIGBUS, local_scan_crash_handler);

  DEBUG(D_receive) debug_printf("calling local_scan(); timeout=%d\n",
    local_scan_timeout);
  local_scan_data = NULL;

  had_local_scan_timeout = 0;
  os_non_restarting_signal(SIGALRM, local_scan_timeout_handler);
  if (local_scan_timeout > 0) ALARM(local_scan_timeout);
  rc = local_scan(data_fd, &local_scan_data);
  ALARM_CLR(0);
  os_non_restarting_signal(SIGALRM, sigalrm_handler);

  f.enable_dollar_recipients = FALSE;

  store_pool = POOL_MAIN;   /* In case changed */
  DEBUG(D_receive) debug_printf("local_scan() returned %d %s\n", rc,
    local_scan_data);

  os_non_restarting_signal(SIGSEGV, SIG_DFL);
  os_non_restarting_signal(SIGFPE, SIG_DFL);
  os_non_restarting_signal(SIGILL, SIG_DFL);
  os_non_restarting_signal(SIGBUS, SIG_DFL);
  }
else
  {
  if (had_local_scan_crash)
    {
    log_write(0, LOG_MAIN|LOG_REJECT, "local_scan() function crashed with "
      "signal %d - message temporarily rejected (size %d)",
      had_local_scan_crash, message_size);
    receive_bomb_out(US"local-scan-error", US"local verification problem");
    /* Does not return */
    }
  if (had_local_scan_timeout)
    {
    log_write(0, LOG_MAIN|LOG_REJECT, "local_scan() function timed out - "
      "message temporarily rejected (size %d)", message_size);
    receive_bomb_out(US"local-scan-timeout", US"local verification problem");
    /* Does not return */
    }
  }

/* The length check is paranoia against some runaway code, and also because
(for a success return) lines in the spool file are read into big_buffer. */

if (local_scan_data)
  {
  int len = Ustrlen(local_scan_data);
  if (len > LOCAL_SCAN_MAX_RETURN) len = LOCAL_SCAN_MAX_RETURN;
  local_scan_data = string_copyn(local_scan_data, len);
  }

if (rc == LOCAL_SCAN_ACCEPT_FREEZE)
  {
  if (!f.deliver_freeze)         /* ACL might have already frozen */
    {
    f.deliver_freeze = TRUE;
    deliver_frozen_at = time(NULL);
    frozen_by = US"local_scan()";
    }
  rc = LOCAL_SCAN_ACCEPT;
  }
else if (rc == LOCAL_SCAN_ACCEPT_QUEUE)
  {
  if (!f.queue_only_policy)      /* ACL might have already queued */
    {
    f.queue_only_policy = TRUE;
    queued_by = US"local_scan()";
    }
  rc = LOCAL_SCAN_ACCEPT;
  }

/* Message accepted: remove newlines in local_scan_data because otherwise
the spool file gets corrupted. Ensure that all recipients are qualified. */

if (rc == LOCAL_SCAN_ACCEPT)
  {
  if (local_scan_data)
    for (uschar * s = local_scan_data; *s; s++) if (*s == '\n') *s = ' ';
  for (recipient_item * r = recipients_list;
       r < recipients_list + recipients_count; r++)
    {
    r->address = rewrite_address_qualify(r->address, TRUE);
    if (r->errors_to)
      r->errors_to = rewrite_address_qualify(r->errors_to, TRUE);
    }
  if (recipients_count == 0 && !blackholed_by)
    blackholed_by = US"local_scan";
  }

/* Message rejected: newlines permitted in local_scan_data to generate
multiline SMTP responses. */

else
  {
  uschar *istemp = US"";
  uschar *smtp_code;
  gstring * g;

  errmsg = local_scan_data;

  Uunlink(spool_name);          /* Cancel this message */
  switch(rc)
    {
    default:
      log_write(0, LOG_MAIN, "invalid return %d from local_scan(). Temporary "
        "rejection given", rc);
      goto TEMPREJECT;

    case LOCAL_SCAN_REJECT_NOLOGHDR:
      BIT_CLEAR(log_selector, log_selector_size, Li_rejected_header);
      /* Fall through */

    case LOCAL_SCAN_REJECT:
      smtp_code = US"550";
      if (!errmsg) errmsg =  US"Administrative prohibition";
      break;

    case LOCAL_SCAN_TEMPREJECT_NOLOGHDR:
      BIT_CLEAR(log_selector, log_selector_size, Li_rejected_header);
      /* Fall through */

    case LOCAL_SCAN_TEMPREJECT:
    TEMPREJECT:
      smtp_code = US"451";
      if (!errmsg) errmsg = US"Temporary local problem";
      istemp = US"temporarily ";
      break;
    }

  g = string_append(NULL, 2, US"F=", *sender_address ? sender_address : US"<>");
  g = add_host_info_for_log(g);

  log_write(0, LOG_MAIN|LOG_REJECT, "%Y %srejected by local_scan(): %.256s",
    g, istemp, string_printing(errmsg));

  if (smtp_input)
    if (!smtp_batched_input)
      {
      smtp_respond(smtp_code, 3, SR_FINAL, errmsg);
      smtp_reply = US"";            /* Indicate reply already sent */
      goto NOT_ACCEPTED;                        /* Skip to end of function */
      }
    else
      moan_smtp_batch(NULL, "%s %s", smtp_code, errmsg);
      /* Does not return */
  else
    {
    fseek(spool_data_file, (long int)spool_data_start_offset(message_id), SEEK_SET);
    give_local_error(ERRMESS_LOCAL_SCAN, errmsg,
      US"message rejected by local scan code: ", error_rc, spool_data_file,
        header_list);
    /* Does not return */
    }
  }

/* Reset signal handlers to ignore signals that previously would have caused
the message to be abandoned. */

signal(SIGTERM, SIG_IGN);
signal(SIGINT, SIG_IGN);
#endif  /* HAVE_LOCAL_SCAN */

/* If we are faking a reject or defer, avoid sennding a DSN for the
actually-accepted message */

if (fake_response != OK)
  for (recipient_item * r = recipients_list;
       r < recipients_list + recipients_count; r++)
    {
    DEBUG(D_receive) if (r->dsn_flags & (rf_notify_success | rf_notify_delay))
      debug_printf("DSN: clearing flags due to fake-response for message\n");
    r->dsn_flags = r->dsn_flags & ~(rf_notify_success | rf_notify_delay)
                    | rf_notify_never;
    }


/* Ensure the first time flag is set in the newly-received message. */

f.deliver_firsttime = TRUE;

#ifdef EXPERIMENTAL_BRIGHTMAIL
if (bmi_run == 1)
  { /* rewind data file */
  lseek(data_fd, (long int)spool_data_start_offset(message_id), SEEK_SET);
  bmi_verdicts = bmi_process_message(header_list, data_fd);
  }
#endif

/* Update the timestamp in our Received: header to account for any time taken by
an ACL or by local_scan(). The new time is the time that all reception
processing is complete. */

timestamp = expand_string(US"${tod_full}");
tslen = Ustrlen(timestamp);

memcpy(received_header->text + received_header->slen - tslen - 1,
  timestamp, tslen);

/* In MUA wrapper mode, ignore queueing actions set by ACL or local_scan() */

if (mua_wrapper)
  {
  f.deliver_freeze = FALSE;
  f.queue_only_policy = FALSE;
  }

/* Keep the data file open until we have written the header file, in order to
hold onto the lock. In a -bh run, or if the message is to be blackholed, we
don't write the header file, and we unlink the data file. If writing the header
file fails, we have failed to accept this message. */

if (host_checking || blackholed_by)
  {
  Uunlink(spool_name);
  msg_size = 0;                                  /* Compute size for log line */
  for (header_line * h = header_list; h; h = h->next)
    if (h->type != '*') msg_size += h->slen;
  }

/* Write the -H file */

else
  if ((msg_size = spool_write_header(message_id, SW_RECEIVING, &errmsg)) < 0)
    {
    log_write(0, LOG_MAIN, "Message abandoned: %s", errmsg);
    Uunlink(spool_name);           /* Lose the data file */

    if (smtp_input)
      {
      smtp_reply = US"451 Error in writing spool file";
      goto NOT_ACCEPTED;
      }
    else
      {
      fseek(spool_data_file, (long int)spool_data_start_offset(message_id), SEEK_SET);
      give_local_error(ERRMESS_IOERR, errmsg, US"", error_rc, spool_data_file,
        header_list);
      /* Does not return */
      }
    }


/* The message has now been successfully received. */

receive_messagecount++;

if (fflush(spool_data_file))
  {
  errmsg = string_sprintf("Spool write error: %s", strerror(errno));
  log_write(0, LOG_MAIN, "%s\n", errmsg);
  Uunlink(spool_name);           /* Lose the data file */

  if (smtp_input)
    {
    smtp_reply = US"451 Error in writing spool file";
    goto NOT_ACCEPTED;
    }
  else
    {
    fseek(spool_data_file, (long int)spool_data_start_offset(message_id), SEEK_SET);
    give_local_error(ERRMESS_IOERR, errmsg, US"", error_rc, spool_data_file,
      header_list);
    /* Does not return */
    }
  }

/* Add data size to written header size. We do not count the initial file name
that is in the file, but we do add one extra for the notional blank line that
precedes the data. This total differs from message_size in that it include the
added Received: header and any other headers that got created locally. */

fstat(data_fd, &statbuf);
msg_size += statbuf.st_size - spool_data_start_offset(message_id) + 1;

/* Generate a "message received" log entry. We do this by building up a dynamic
string as required.  We log the arrival of a new message while the
file is still locked, just in case the machine is *really* fast, and delivers
it first! Include any message id that is in the message - since the syntax of a
message id is actually an addr-spec, we can use the parse routine to canonicalize
it. */

rcvd_log_reset_point = store_mark();
g = string_get(256);

g = string_append(g, 2,
  fake_response == FAIL ? US"(= " : US"<= ",
  *sender_address ? sender_address : US"<>");
if (message_reference)
  g = string_append(g, 2, US" R=", message_reference);

g = add_host_info_for_log(g);

#ifndef DISABLE_TLS
if (LOGGING(tls_cipher) && tls_in.cipher)
  {
  g = string_append(g, 2, US" X=", tls_in.cipher);
# ifndef DISABLE_TLS_RESUME
  if (LOGGING(tls_resumption) && tls_in.resumption & RESUME_USED)
    g = string_catn(g, US"*", 1);
# endif
  }
if (LOGGING(tls_certificate_verified) && tls_in.cipher)
  g = string_append(g, 2, US" CV=", tls_in.certificate_verified ? "yes":"no");
if (LOGGING(tls_peerdn) && tls_in.peerdn)
  g = string_append(g, 3, US" DN=\"", string_printing(tls_in.peerdn), US"\"");
if (LOGGING(tls_sni) && tls_in.sni)
  g = string_append(g, 2, US" SNI=", string_printing2(tls_in.sni, SP_TAB|SP_SPACE));
#endif

if (sender_host_authenticated)
  {
  g = string_append(g, 2, US" A=", sender_host_authenticated);
  if (authenticated_id)
    {
    g = string_append(g, 2, US":", authenticated_id);
    if (LOGGING(smtp_mailauth) && authenticated_sender)
      g = string_append(g, 2, US":", authenticated_sender);
    }
  }

#ifndef DISABLE_PRDR
if (prdr_requested)
  g = string_catn(g, US" PRDR", 5);
#endif

#ifdef SUPPORT_PROXY
if (proxy_session && LOGGING(proxy))
  g = string_append(g, 2, US" PRX=", proxy_local_address);
#endif

if (chunking_state > CHUNKING_OFFERED)
  g = string_catn(g, US" K", 2);

g = string_fmt_append(g, " S=%d", msg_size);

/* log 8BITMIME mode announced in MAIL_FROM
   0 ... no BODY= used
   7 ... 7BIT
   8 ... 8BITMIME */
if (LOGGING(8bitmime))
  g = string_fmt_append(g, " M8S=%d", body_8bitmime);

#ifndef DISABLE_DKIM
if (LOGGING(dkim) && dkim_verify_overall)
  g = string_append(g, 2, US" DKIM=", dkim_verify_overall);
# ifdef EXPERIMENTAL_ARC
if (LOGGING(dkim) && arc_state && Ustrcmp(arc_state, "pass") == 0)
  g = string_catn(g, US" ARC", 4);
# endif
#endif

if (LOGGING(receive_time))
  {
  struct timeval diff = received_time_complete;
  timediff(&diff, &received_time);
  g = string_append(g, 2, US" RT=", string_timediff(&diff));
  }

if (*queue_name)
  g = string_append(g, 2, US" Q=", queue_name);

/* If an addr-spec in a message-id contains a quoted string, it can contain
any characters except " \ and CR and so in particular it can contain NL!
Therefore, make sure we use a printing-characters only version for the log.
Also, allow for domain literals in the message id. */

if (  LOGGING(msg_id) && msgid_header
   && (LOGGING(msg_id_created) || !msgid_header_newly_created)
   )
  {
  uschar * old_id;
  BOOL save_allow_domain_literals = allow_domain_literals;
  allow_domain_literals = TRUE;
  int start, end, domain;

  old_id = parse_extract_address(Ustrchr(msgid_header->text, ':') + 1,
    &errmsg, &start, &end, &domain, FALSE);
  allow_domain_literals = save_allow_domain_literals;
  if (old_id)
    g = string_append(g, 2,
      msgid_header_newly_created ? US" id*=" : US" id=",
      string_printing(old_id));
  }

/* If subject logging is turned on, create suitable printing-character
text. By expanding $h_subject: we make use of the MIME decoding. */

if (LOGGING(subject) && subject_header)
  {
  uschar *p = big_buffer;
  uschar *ss = expand_string(US"$h_subject:");

  /* Backslash-quote any double quotes or backslashes so as to make a
  a C-like string, and turn any non-printers into escape sequences. */

  *p++ = '\"';
  if (*ss != 0) for (int i = 0; i < 100 && ss[i] != 0; i++)
    {
    if (ss[i] == '\"' || ss[i] == '\\') *p++ = '\\';
    *p++ = ss[i];
    }
  *p++ = '\"';
  *p = 0;
  g = string_append(g, 2, US" T=", string_printing(big_buffer));
  }

/* Terminate the string: string_cat() and string_append() leave room, but do
not put the zero in. */

(void) string_from_gstring(g);

/* Create a message log file if message logs are being used and this message is
not blackholed. Write the reception stuff to it. We used to leave message log
creation until the first delivery, but this has proved confusing for some
people. */

if (message_logs && !blackholed_by)
  {
  int fd;
  uschar * m_name = spool_fname(US"msglog", message_subdir, message_id, US"");

  if (  (fd = Uopen(m_name, O_WRONLY|O_APPEND|O_CREAT, SPOOL_MODE)) < 0
     && errno == ENOENT
     )
    {
    (void)directory_make(spool_directory,
                        spool_sname(US"msglog", message_subdir),
                        MSGLOG_DIRECTORY_MODE, TRUE);
    fd = Uopen(m_name, O_WRONLY|O_APPEND|O_CREAT, SPOOL_MODE);
    }

  if (fd < 0)
    log_write(0, LOG_MAIN|LOG_PANIC, "Couldn't open message log %s: %s",
      m_name, strerror(errno));
  else
    {
    FILE *message_log = fdopen(fd, "a");
    if (!message_log)
      {
      log_write(0, LOG_MAIN|LOG_PANIC, "Couldn't fdopen message log %s: %s",
        m_name, strerror(errno));
      (void)close(fd);
      }
    else
      {
      uschar * now = tod_stamp(tod_log);
      /* Drop the initial "<= " */
      fprintf(message_log, "%s Received from %s\n", now, g->s+3);
      if (f.deliver_freeze) fprintf(message_log, "%s frozen by %s\n", now,
        frozen_by);
      if (f.queue_only_policy) fprintf(message_log,
        "%s no immediate delivery: queued%s%s by %s\n", now,
        *queue_name ? " in " : "", *queue_name ? CS queue_name : "",
        queued_by);
      (void)fclose(message_log);
      }
    }
  }

/* Everything has now been done for a successful message except logging its
arrival, and outputting an SMTP response. While writing to the log, set a flag
to cause a call to receive_bomb_out() if the log cannot be opened. */

f.receive_call_bombout = TRUE;

/* Before sending an SMTP response in a TCP/IP session, we check to see if the
connection has gone away. This can only be done if there is no unconsumed input
waiting in the local input buffer. We can test for this by calling
receive_hasc(). RFC 2920 (pipelining) explicitly allows for additional
input to be sent following the final dot, so the presence of following input is
not an error.

If the connection is still present, but there is no unread input for the
socket, the result of a select() call will be zero. If, however, the connection
has gone away, or if there is pending input, the result of select() will be
non-zero. The two cases can be distinguished by trying to read the next input
character. If we succeed, we can unread it so that it remains in the local
buffer for handling later. If not, the connection has been lost.

Of course, since TCP/IP is asynchronous, there is always a chance that the
connection will vanish between the time of this test and the sending of the
response, but the chance of this happening should be small. */

if (  smtp_input && sender_host_address && !f.sender_host_notsocket
   && !receive_hasc())
  {
  if (poll_one_fd(fileno(smtp_in), POLLIN, 0) != 0)
    {
    int c = (receive_getc)(GETC_BUFFER_UNLIMITED);
    if (c != EOF) (receive_ungetc)(c);
    else
      {
      smtp_notquit_exit(US"connection-lost", NULL, NULL);
      smtp_reply = US"";    /* No attempt to send a response */
      smtp_yield = FALSE;   /* Nothing more on this connection */

      /* Re-use the log line workspace */

      gstring_reset(g);
      g = string_cat(g, US"SMTP connection lost after final dot");
      g = add_host_info_for_log(g);
      log_write(0, LOG_MAIN, "%Y", g);

      /* Delete the files for this aborted message. */

      Uunlink(spool_name);
      Uunlink(spool_fname(US"input", message_subdir, message_id, US"-H"));
      Uunlink(spool_fname(US"msglog", message_subdir, message_id, US""));

      goto TIDYUP;
      }
    }
  }

/* The connection has not gone away; we really are going to take responsibility
for this message. */

/* Cutthrough - had sender last-dot; assume we've sent (or bufferred) all
   data onward by now.

   Send dot onward.  If accepted, wipe the spooled files, log as delivered and accept
   the sender's dot (below).
   If rejected: copy response to sender, wipe the spooled files, log appropriately.
   If temp-reject: normally accept to sender, keep the spooled file - unless defer=pass
   in which case pass temp-reject back to initiator and dump the files.

   Having the normal spool files lets us do data-filtering, and store/forward on temp-reject.

   XXX We do not handle queue-only, freezing, or blackholes.
*/
if(cutthrough.cctx.sock >= 0 && cutthrough.delivery)
  {
  uschar * msg = cutthrough_finaldot(); /* Ask the target system to accept the message */
                                        /* Logging was done in finaldot() */
  switch(msg[0])
    {
    case '2':   /* Accept. Do the same to the source; dump any spoolfiles.   */
      cutthrough_done = ACCEPTED;
      break;                                    /* message_id needed for SMTP accept below */

    case '4':   /* Temp-reject. Keep spoolfiles and accept, unless defer-pass mode.
                ... for which, pass back the exact error */
      if (cutthrough.defer_pass) smtp_reply = string_copy_perm(msg, TRUE);
      cutthrough_done = TMP_REJ;                /* Avoid the usual immediate delivery attempt */
      break;                                    /* message_id needed for SMTP accept below */

    default:    /* Unknown response, or error.  Treat as temp-reject.         */
      if (cutthrough.defer_pass) smtp_reply = US"450 Onward transmission not accepted";
      cutthrough_done = TMP_REJ;                /* Avoid the usual immediate delivery attempt */
      break;                                    /* message_id needed for SMTP accept below */

    case '5':   /* Perm-reject.  Do the same to the source.  Dump any spoolfiles */
      smtp_reply = string_copy_perm(msg, TRUE);         /* Pass on the exact error */
      cutthrough_done = PERM_REJ;
      break;
    }
  }

#ifndef DISABLE_PRDR
if(!smtp_reply || prdr_requested)
#else
if(!smtp_reply)
#endif
  {
  log_write(0, LOG_MAIN |
                (LOGGING(received_recipients) ? LOG_RECIPIENTS : 0) |
                (LOGGING(received_sender) ? LOG_SENDER : 0),
            "%Y", g);

  /* Log any control actions taken by an ACL or local_scan(). */

  if (f.deliver_freeze) log_write(0, LOG_MAIN, "frozen by %s", frozen_by);
  if (f.queue_only_policy) log_write(L_delay_delivery, LOG_MAIN,
    "no immediate delivery: queued%s%s by %s",
    *queue_name ? " in " : "", *queue_name ? CS queue_name : "",
    queued_by);
  }
f.receive_call_bombout = FALSE;

/* The store for the main log message can be reused */
rcvd_log_reset_point = store_reset(rcvd_log_reset_point);

/* If the message is frozen, and freeze_tell is set, do the telling. */

if (f.deliver_freeze && freeze_tell && freeze_tell[0])
  moan_tell_someone(freeze_tell, NULL, US"Message frozen on arrival",
    "Message %s was frozen on arrival by %s.\nThe sender is <%s>.\n",
    message_id, frozen_by, sender_address);


/* Either a message has been successfully received and written to the two spool
files, or an error in writing the spool has occurred for an SMTP message, or
an SMTP message has been rejected for policy reasons, or a message was passed on
by cutthrough delivery. (For a non-SMTP message we will have already given up
because there's no point in carrying on!) For non-cutthrough we must now close
(and thereby unlock) the data file. In the successful case, this leaves the
message on the spool, ready for delivery. In the error case, the spool file will
be deleted. Then tidy up store, interact with an SMTP call if necessary, and
return.

For cutthrough we hold the data file locked until we have deleted it, otherwise
a queue-runner could grab it in the window.

A fflush() was done earlier in the expectation that any write errors on the
data file will be flushed(!) out thereby. Nevertheless, it is theoretically
possible for fclose() to fail - and this has been seen on obscure filesystems
(probably one that delayed the actual media write as long as possible)
but what to do? What has happened to the lock if this happens?
It's a mess because we already logged the acceptance.
We can at least log the issue, try to remove spoolfiles and respond with
a temp-reject.  We do not want to close before logging acceptance because
we want to hold the lock until we know that logging worked.
Could we make this less likely by doing an fdatasync() just after the fflush()?
That seems like a good thing on data-security grounds, but how much will it hit
performance? */


goto TIDYUP;

NOT_ACCEPTED:
message_id[0] = 0;                              /* Indicate no message accepted */

TIDYUP:
process_info[process_info_len] = 0;                     /* Remove message id */
if (spool_data_file && cutthrough_done == NOT_TRIED)
  {
  if (fclose(spool_data_file))                          /* Frees the lock */
    {
    log_msg = string_sprintf("spoolfile error on close: %s", strerror(errno));
    log_write(0, LOG_MAIN|LOG_PANIC |
                  (LOGGING(received_recipients) ? LOG_RECIPIENTS : 0) |
                  (LOGGING(received_sender) ? LOG_SENDER : 0),
              "%s", log_msg);
    log_write(0, LOG_MAIN |
                  (LOGGING(received_recipients) ? LOG_RECIPIENTS : 0) |
                  (LOGGING(received_sender) ? LOG_SENDER : 0),
              "rescind the above message-accept");

    Uunlink(spool_name);
    Uunlink(spool_fname(US"input", message_subdir, message_id, US"-H"));
    Uunlink(spool_fname(US"msglog", message_subdir, message_id, US""));

    /* Claim a data ACL temp-reject, just to get reject logging and response */
    if (smtp_input) smtp_handle_acl_fail(ACL_WHERE_DATA, rc, NULL, log_msg);
    smtp_reply = US"";          /* Indicate reply already sent */

    message_id[0] = 0;          /* no message accepted */
    }
  spool_data_file = NULL;
  }

/* Now reset signal handlers to their defaults */

signal(SIGTERM, SIG_DFL);
signal(SIGINT, SIG_DFL);

/* Tell an SMTP caller the state of play, and arrange to return the SMTP return
value, which defaults TRUE - meaning there may be more incoming messages from
this connection. For non-SMTP callers (where there is only ever one message),
the default is FALSE. */

if (smtp_input)
  {
  yield = smtp_yield;

  /* Handle interactive SMTP callers. After several kinds of error, smtp_reply
  is set to the response that should be sent. When it is NULL, we generate
  default responses. After an ACL error or local_scan() error, the response has
  already been sent, and smtp_reply is an empty string to indicate this. */

  if (!smtp_batched_input)
    {
    if (!smtp_reply)
      {
      if (fake_response != OK)
        smtp_respond(fake_response == DEFER ? US"450" : US"550",
          3, SR_FINAL, fake_response_text);

      /* An OK response is required; use "message" text if present. */

      else if (user_msg)
        {
        uschar *code = US"250";
        int len = 3;
        smtp_message_code(&code, &len, &user_msg, NULL, TRUE);
        smtp_respond(code, len, SR_FINAL, user_msg);
        }

      /* Default OK response */

      else if (chunking_state > CHUNKING_OFFERED)
        {
        /* If there is more input waiting, no need to flush (probably the client
        pipelined QUIT after data).  We check only the in-process buffer, not
        the socket. */

        smtp_printf("250- %u byte chunk, total %d\r\n250 OK id=%s\r\n",
            receive_hasc(),
            chunking_datasize, message_size+message_linecount, message_id);
        chunking_state = CHUNKING_OFFERED;
        }
      else
        smtp_printf("250 OK id=%s\r\n", receive_hasc(), message_id);

      if (host_checking)
        fprintf(stdout,
          "\n**** SMTP testing: that is not a real message id!\n\n");
      }

    /* smtp_reply is set non-empty */

    else if (smtp_reply[0] != 0)
      if (fake_response != OK && smtp_reply[0] == '2')
        smtp_respond(fake_response == DEFER ? US"450" : US"550",
                      3, SR_FINAL, fake_response_text);
      else
        smtp_printf("%.1024s\r\n", SP_NO_MORE, smtp_reply);

    switch (cutthrough_done)
      {
      case ACCEPTED:
        log_write(0, LOG_MAIN, "Completed");/* Delivery was done */
      case PERM_REJ:
                                                         /* Delete spool files */
        Uunlink(spool_name);
        Uunlink(spool_fname(US"input", message_subdir, message_id, US"-H"));
        Uunlink(spool_fname(US"msglog", message_subdir, message_id, US""));
        break;

      case TMP_REJ:
        if (cutthrough.defer_pass)
          {
          Uunlink(spool_name);
          Uunlink(spool_fname(US"input", message_subdir, message_id, US"-H"));
          Uunlink(spool_fname(US"msglog", message_subdir, message_id, US""));
          }
      default:
        break;
      }
    if (cutthrough_done != NOT_TRIED)
      {
      if (spool_data_file)
        {
        (void) fclose(spool_data_file);  /* Frees the lock; do not care if error */
        spool_data_file = NULL;
        }
      message_id[0] = 0;          /* Prevent a delivery from starting */
      cutthrough.delivery = cutthrough.callout_hold_only = FALSE;
      cutthrough.defer_pass = FALSE;
      }
    }

  /* For batched SMTP, generate an error message on failure, and do
  nothing on success. The function moan_smtp_batch() does not return -
  it exits from the program with a non-zero return code. */

  else if (smtp_reply)
    moan_smtp_batch(NULL, "%s", smtp_reply);
  }


/* If blackholing, we can immediately log this message's sad fate. The data
file has already been unlinked, and the header file was never written to disk.
We must now indicate that nothing was received, to prevent a delivery from
starting. */

if (blackholed_by)
  {
  const uschar *detail =
#ifdef HAVE_LOCAL_SCAN
    local_scan_data ? string_printing(local_scan_data) :
#endif
    string_sprintf("(%s discarded recipients)", blackholed_by);
  log_write(0, LOG_MAIN, "=> blackhole %s%s", detail, blackhole_log_msg);
  log_write(0, LOG_MAIN, "Completed");
  message_id[0] = 0;
  }

/* Reset headers so that logging of rejects for a subsequent message doesn't
include them. It is also important to set header_last = NULL before exiting
from this function, as this prevents certain rewrites that might happen during
subsequent verifying (of another incoming message) from trying to add headers
when they shouldn't. */

header_list = header_last = NULL;

return yield;  /* TRUE if more messages (SMTP only) */
}

/* End of receive.c */