flocksim.c

广泛使用的邮件服务器！同时

/* ========================================================================
 * Copyright 1988-2007 University of Washington
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * 
 * ========================================================================
 */

/*
 * Program:	flock emulation via fcntl() locking
 *
 * Author:	Mark Crispin
 *		Networks and Distributed Computing
 *		Computing & Communications
 *		University of Washington
 *		Administration Building, AG-44
 *		Seattle, WA  98195
 *		Internet: [e-mail]
 *
 * Date:	10 April 2001
 * Last Edited:	11 October 2007
 */
 
#undef flock			/* name is used as a struct for fcntl */

#ifndef NOFSTATVFS		/* thank you, SUN.  NOT! */
# ifndef NOFSTATVFS64
#  ifndef _LARGEFILE64_SOURCE
#   define _LARGEFILE64_SOURCE
#  endif	/* _LARGEFILE64_SOURCE */
# endif		/* NOFSTATVFFS64 */
#include <sys/statvfs.h>
#endif		/* NOFSTATVFS */

#ifndef NSIG			/* don't know if this can happen */
#define NSIG 32			/* a common maximum */
#endif

/* Emulator for flock() call
 * Accepts: file descriptor
 *	    operation bitmask
 * Returns: 0 if successful, -1 if failure under BSD conditions
 */

int flocksim (int fd,int op)
{
  char tmp[MAILTMPLEN];
  int logged = 0;
  struct stat sbuf;
  struct ustat usbuf;
  struct flock fl;
				/* lock zero bytes at byte 0 */
  fl.l_whence = SEEK_SET; fl.l_start = fl.l_len = 0;
  fl.l_pid = getpid ();		/* shouldn't be necessary */
  switch (op & ~LOCK_NB) {	/* translate to fcntl() operation */
  case LOCK_EX:			/* exclusive */
    fl.l_type = F_WRLCK;
    break;
  case LOCK_SH:			/* shared */
    fl.l_type = F_RDLCK;
    break;
  case LOCK_UN:			/* unlock */
    fl.l_type = F_UNLCK;
    break;
  default:			/* default */
    errno = EINVAL;
    return -1;
  }
				/* always return success if disabled */
  if (mail_parameters (NIL,GET_DISABLEFCNTLLOCK,NIL)) return 0;

  /*  Make fcntl() locking of NFS files be a no-op the way it is with flock()
   * on BSD.  This is because the rpc.statd/rpc.lockd daemons don't work very
   * well and cause cluster-wide hangs if you exercise them at all.  The
   * result of this is that you lose the ability to detect shared mail_open()
   * on NFS-mounted files.  If you are wise, you'll use IMAP instead of NFS
   * for mail files.
   *
   *  Sun alleges that it doesn't matter, and that they have fixed all the
   * rpc.statd/rpc.lockd bugs.  As of October 2006, that is still false.
   *
   *  We need three tests for three major historical variants in SVR4:
   *  1) In NFSv2, ustat() would return -1 in f_tinode for NFS.
   *  2) When fstatvfs() was introduced with NFSv3, ustat() was "fixed".
   *  3) When 64-bit filesystems were introduced, fstatvfs() would return
   *	 EOVERFLOW; you have to use fstatvfs64() even though you don't care
   *	 about any of the affected values.
   *
   * We can't use fstatfs() because fstatfs():
   * . is documented as being deprecated in SVR4.
   * . has inconsistent calling conventions (there are two additional int
   *   arguments on Solaris and I don't know what they do).
   * . returns inconsistent statfs structs.  On Solaris, the file system type
   *   is a short called f_fstyp.  On AIX, it's an int called f_type that is
   *   documented as always being 0!
   *
   * For what it's worth, here's the scoop on fstatfs() elsewhere:
   *
   *  On Linux, the file system type is a long called f_type that has a file
   * system type code.  A different module (flocklnx.c) uses this because
   * some knothead "improved" flock() to return ENOLCK on NFS files instead
   * of being a successful no-op.  This "improvement" apparently has been
   * reverted, but not before it got to many systems in the field.
   *
   *  On BSD, it's a short called either f_otype or f_type that is documented
   * as always being zero.  Fortunately, BSD has flock() the way it's supposed
   * to be, and none of this nonsense is necessary.
   */
  if (!fstat (fd,&sbuf))	{ /* no hope of working if can't fstat()! */
    /* Any base type that begins with "nfs" or "afs" is considered to be a
     * network filesystem.
     */
#ifndef NOFSTATVFS
    struct statvfs vsbuf;
#ifndef NOFSTATVFS64
    struct statvfs64 vsbuf64;
    if (!fstatvfs64 (fd,&vsbuf64) && (vsbuf64.f_basetype[1] == 'f') &&
	(vsbuf64.f_basetype[2] == 's') &&
	((vsbuf64.f_basetype[0] == 'n') || (vsbuf64.f_basetype[0] == 'a')))
      return 0;
#endif		/* NOFSTATVFS64 */
    if (!fstatvfs (fd,&vsbuf) && (vsbuf.f_basetype[1] == 'f') &&
	(vsbuf.f_basetype[2] == 's') &&
	((vsbuf.f_basetype[0] == 'n') || (vsbuf.f_basetype[0] == 'a')))
      return 0;
#endif		/* NOFSTATVFS */
    if (!ustat (sbuf.st_dev,&usbuf) && !++usbuf.f_tinode) return 0;
  }

				/* do the lock */
  while (fcntl (fd,(op & LOCK_NB) ? F_SETLK : F_SETLKW,&fl))
    if (errno != EINTR) {
      /* Can't use switch here because these error codes may resolve to the
       * same value on some systems.
       */
      if ((errno != EWOULDBLOCK) && (errno != EAGAIN) && (errno != EACCES)) {
	sprintf (tmp,"Unexpected file locking failure: %.100s",
		 strerror (errno));
				/* give the user a warning of what happened */
	MM_NOTIFY (NIL,tmp,WARN);
	if (!logged++) syslog (LOG_ERR,"%s",tmp);
	if (op & LOCK_NB) return -1;
	sleep (5);		/* slow things down for loops */
      }
				/* return failure for non-blocking lock */
      else if (op & LOCK_NB) return -1;
    }
  return 0;			/* success */
}

/* Master/slave procedures for safe fcntl() locking.
 *
 *  The purpose of this nonsense is to work around a bad bug in fcntl()
 * locking.  The cretins who designed it decided that a close() should
 * release any locks made by that process on the file opened on that
 * file descriptor.  Never mind that the lock wasn't made on that file
 * descriptor, but rather on some other file descriptor.
 *
 *  This bug is on every implementation of fcntl() locking that I have
 * tested.  Fortunately, on BSD systems, OSF/1, and Linux, we can use the
 * flock() system call which doesn't have this bug.
 *
 *  Note that OSF/1, Linux, and some BSD systems have both broken fcntl()
 * locking and the working flock() locking.
 *
 *  The program below can be used to demonstrate this problem.  Be sure to
 * let it run long enough for all the sleep() calls to finish.
 */

#if 0
#include <stdio.h>
#include <unistd.h>
#include <fcntl.h>
#include <errno.h>
#include <string.h>
#include <sys/file.h>

main ()
{
  struct flock fl;
  int fd,fd2;
  char *file = "a.a";
  if ((fd = creat (file,0666)) < 0)
    perror ("TEST FAILED: can't create test file"),_exit (errno);
  close (fd);
  if (fork ()) {		/* parent */
    if ((fd = open (file,O_RDWR,0)) < 0) abort();
				/* lock applies to entire file */
    fl.l_whence = fl.l_start = fl.l_len = 0;
    fl.l_pid = getpid ();	/* shouldn't be necessary */
    fl.l_type = F_RDLCK;
    if (fcntl (fd,F_SETLKW,&fl) == -1) abort ();
    sleep (5);
    if ((fd2 = open (file,O_RDWR,0)) < 0) abort ();
    sleep (1);
    puts ("parent test ready -- will hang here if locking works correctly");
    close (fd2);
    wait (0);
    puts ("OS BUG: child terminated");
    _exit (0);
  }
  else {			/* child */
    sleep (2);
    if ((fd = open (file,O_RDWR,0666)) < 0) abort ();
    puts ("child test ready -- child will hang if no bug");
				/* lock applies to entire file */
    fl.l_whence = fl.l_start = fl.l_len = 0;
    fl.l_pid = getpid ();	/* shouldn't be necessary */
    fl.l_type = F_WRLCK;
    if (fcntl (fd,F_SETLKW,&fl) == -1) abort ();
    puts ("OS BUG: child got lock");
  }
}
#endif

/*  Beware of systems such as AIX which offer flock() as a compatibility
 * function that is just a jacket into fcntl() locking.  The program below
 * is a variant of the program above, only using flock().  It can be used
 * to test to see if your system has real flock() or just a jacket into
 * fcntl().
 *
 *  Be sure to let it run long enough for all the sleep() calls to finish.
 * If the program hangs, then flock() works and you can dispense with the
 * use of this module (you lucky person!).
 */

#if 0
#include <stdio.h>
#include <errno.h>
#include <string.h>
#include <sys/file.h>

main ()
{
  int fd,fd2;
  char *file = "a.a";
  if ((fd = creat (file,0666)) < 0)
    perror ("TEST FAILED: can't create test file"),_exit (errno);
  close (fd);
  if (fork ()) {		/* parent */
    if ((fd = open (file,O_RDWR,0)) < 0) abort();
    if (flock (fd,LOCK_SH) == -1) abort ();
    sleep (5);
    if ((fd2 = open (file,O_RDWR,0)) < 0) abort ();
    sleep (1);
    puts ("parent test ready -- will hang here if flock() works correctly");
    close (fd2);
    wait (0);
    puts ("OS BUG: child terminated");
    _exit (0);
  }
  else {			/* child */
    sleep (2);
    if ((fd = open (file,O_RDWR,0666)) < 0) abort ();
    puts ("child test ready -- child will hang if no bug");
    if (flock (fd,LOCK_EX) == -1) abort ();
    puts ("OS BUG: child got lock");
  }
}
#endif

/* Master/slave details
 *
 *  On broken systems, we invoke an inferior fork to execute any driver
 * dispatches which are likely to tickle this bug; specifically, any
 * dispatch which may fiddle with a mailbox that is already selected.  As
 * of this writing, these are: delete, rename, status, scan, copy, and append.
 *
 *  Delete and rename are pretty marginal, yet there are certain clients
 * (e.g. Outlook Express) that really want to delete or rename the selected
 * mailbox.  The same is true of status, but there are people (such as the
 * authors of Entourage) who don't understand why status of the selected
 * mailbox is bad news.
 *
 *  However, in copy and append it is reasonable to do this to a selected
 * mailbox.  Although scanning the selected mailbox isn't particularly
 * sensible, it's hard to avoid due to wildcards.
 *
 *  It is still possible for an application to trigger the bug by doing
 * mail_open() on the same mailbox twice.  Don't do it.
 *
 *  Once the slave is invoked, the master only has to read events from the
 * slave's output (see below for these events) and translate these events
 * to the appropriate c-client callback.  When end of file occurs on the pipe,
 * the master reads the slave's exit status and uses that as the function
 * return.  The append master is slightly more complicated because it has to
 * send data back to the slave (see below).
 *
 *  The slave takes callback events from the driver which otherwise would
 * pass to the main program.  Only those events which a slave can actually
 * encounter are covered here; for example mm_searched() and mm_list() are
 * not covered since a slave never does the operations that trigger these.
 * Certain other events (mm_exists(), mm_expunged(), mm_flags()) are discarded
 * by the slave since the master will generate these events for itself.
 *
 *  The other events cause the slave to write a newline-terminated string to
 * its output.  The first character of string indicates the event: S for
 * mm_status(), N for mm_notify(), L for mm_log(), C for mm_critical(), X for
 * mm_nocritical(), D for mm_diskerror(), F for mm_fatal(), and "A" for append
 * argument callback.  Most of these events also carry data, which carried as
 * text space-delimited in the string.
 *
 *  Append argument callback requires the master to provide the slave with
 * data in the slave's input.  The first thing that the master provides is
 * either a "+" (master has data for the slave) or a "-" (master has no data).
 * If the master has data, it will then send the flags, internal date, and
 * message text, each as <text octet count><SPACE><text>.
 */

/*  It should be alright for lockslavep to be a global, since it will always
 * be zero in the master (which is where threads would be).  The slave won't
 * ever thread, since any driver which threads in its methods probably can't
 * use fcntl() locking so won't have DR_LOCKING in its driver flags 
 *
 *  lockslavep can not be a static, since it's used by the dispatch macros.
 */

int lockslavep = 0;		/* non-zero means slave process for locking */
static int lockproxycopy = 0;	/* non-zero means redo copy as proxy */
FILE *slavein = NIL;		/* slave input */
FILE *slaveout = NIL;		/* slave output */


/* Common master
 * Accepts: permitted stream
 *	    append callback (append calls only, else NIL)
 *	    data for callback (append calls only, else NIL)
 * Returns: (master) T if slave succeeded, NIL if slave failed
 *	    (slave) NIL always, with lockslavep non-NIL
 */

static long master (MAILSTREAM *stream,append_t af,void *data)
{
  MAILSTREAM *st;
  MAILSTATUS status;
  STRING *message;
  FILE *pi,*po;
  blocknotify_t bn = (blocknotify_t) mail_parameters (NIL,GET_BLOCKNOTIFY,NIL);
  long ret = NIL;
  unsigned long i,j;
  int c,pid,pipei[2],pipeo[2];
  char *s,*t,event[MAILTMPLEN],tmp[MAILTMPLEN];
  lockproxycopy = NIL;		/* not doing a lock proxycopy */
				/* make pipe from slave */
  if (pipe (pipei) < 0) mm_log ("Can't create input pipe",ERROR);
  else if (pipe (pipeo) < 0) {
    mm_log ("Can't create output pipe",ERROR);
    close (pipei[0]); close (pipei[1]);
  }
  else if ((pid = fork ()) < 0) {/* make slave */
    mm_log ("Can't create execution process",ERROR);
    close (pipei[0]); close (pipei[1]);
    close (pipeo[0]); close (pipeo[1]);
  }
  else if (lockslavep = !pid) {	/* are we slave or master? */
    alarm (0);			/* slave doesn't have alarms or signals */
    for (c = 0; c < NSIG; c++) signal (c,SIG_DFL);
    if (!(slavein = fdopen (pipeo[0],"r")) ||
	!(slaveout = fdopen (pipei[1],"w")))
      fatal ("Can't do slave pipe buffered I/O");
    close (pipei[0]);		/* close parent's side of the pipes */
    close (pipeo[1]);
  }

  else {			/* master process */
    void *blockdata = (*bn) (BLOCK_SENSITIVE,NIL);
    close (pipei[1]);		/* close slave's side of the pipes */
    close (pipeo[0]);
    if (!(pi = fdopen (pipei[0],"r")) || !(po = fdopen (pipeo[1],"w")))
      fatal ("Can't do master pipe buffered I/O");
				/* do slave events until EOF */
				/* read event */
    while (fgets (event,MAILTMPLEN-1,pi)) {
      if (!(s = strchr (event,'\n'))) {
	sprintf (tmp,"Execution process event string too long: %.500s",event);
	fatal (tmp);
      }
      *s = '\0';		/* tie off event at end of line */
      switch (event[0]) {	/* analyze event */
      case 'A':			/* append callback */
	if ((*af) (NIL,data,&s,&t,&message)) {
	  if (i = message ? SIZE (message) : 0) {
	    if (!s) s = "";	/* default values */
	    if (!t) t = "";
	  }
	  else s = t = "";	/* no flags or date if no message */
	  errno = NIL;		/* reset last error */
				/* build response */
	  if (fprintf (po,"+%lu %s%lu %s%lu ",strlen (s),s,strlen (t),t,i) < 0)
	    fatal ("Failed to pipe append command");
				/* write message text */
	  if (i) do if (putc (c = 0xff & SNX (message),po) == EOF) {
	    sprintf (tmp,"Failed to pipe %lu bytes (of %lu), last=%u: %.100s",
		     i,message->size,c,strerror (errno));
	    fatal (tmp);
	  } while (--i);
	}
	else putc ('-',po);	/* append error */
	fflush (po);
	break;
      case '&':			/* slave wants a proxycopy? */
	lockproxycopy = T;
	break;

      case 'L':			/* mm_log() */
	i = strtoul (event+1,&s,10);
	if (!s || (*s++ != ' ')) {
	  sprintf (tmp,"Invalid log event arguments: %.500s",event);
	  fatal (tmp);
	}
	mm_log (s,i);
	break;
      case 'N':			/* mm_notify() */
	st = (MAILSTREAM *) strtoul (event+1,&s,16);
	if (s && (*s++ == ' ')) {
	  i = strtoul (s,&s,10);/* get severity */
	  if (s && (*s++ == ' ')) {
	    mm_notify ((st == stream) ? stream : NIL,s,i);
	    break;
	  }
	}
	sprintf (tmp,"Invalid notify event arguments: %.500s",event);
	fatal (tmp);

      case 'S':			/* mm_status() */
	st = (MAILSTREAM *) strtoul (event+1,&s,16);
	if (s && (*s++ == ' ')) {
	  status.flags = strtoul (s,&s,10);
	  if (s && (*s++ == ' ')) {
	    status.messages = strtoul (s,&s,10);
	    if (s && (*s++ == ' ')) {
	      status.recent = strtoul (s,&s,10);
	      if (s && (*s++ == ' ')) {
		status.unseen = strtoul (s,&s,10);
		if (s && (*s++ == ' ')) {
		  status.uidnext = strtoul (s,&s,10);
		  if (s && (*s++ == ' ')) {
		    status.uidvalidity = strtoul (s,&s,10);
		    if (s && (*s++ == ' ')) {
		      mm_status ((st == stream) ? stream : NIL,s,&status);
		      break;
		    }
		  }
		}
	      }
	    }
	  }
	}
	sprintf (tmp,"Invalid status event arguments: %.500s",event);
	fatal (tmp);
      case 'C':			/* mm_critical() */
	st = (MAILSTREAM *) strtoul (event+1,&s,16);
	mm_critical ((st == stream) ? stream : NIL);
	break;
      case 'X':			/* mm_nocritical() */
	st = (MAILSTREAM *) strtoul (event+1,&s,16);