1. General/Software Feature Questions

1.1 Can I set up a POP or IMAP server on UNIX/Linux/OSF/etc.?

Yes. Refer to the UNIX specific notes in files CONFIG and BUILD.

1.2 I am currently using qpopper as my POP3 server on UNIX. Do I need to replace it with ipop3d in order to run imapd?

Not necessarily.

Although ipop3d interoperates with imapd better than qpopper, imapd and qpopper will work together. The few qpopper/imapd interoperability issues mostly affect users who use both IMAP and POP3 clients; those users would probably be better served if their POP3 server is ipop3d.

If you are happy with qpopper and just want to add imapd, you should do that, and defer a decision on changing qpopper to ipop3d. That way, you can get comfortable with imapd's performance, without changing anything for your qpopper users.

Many sites have subsequently decided to change from qpopper to ipop3d in order to get better POP3/IMAP interoperability. If you need to do this, you'll know. There also seems to be a way to make qpopper work better with imapd; see the answer to the My qpopper users keep on getting the DON'T DELETE THIS MESSAGE -- FOLDER INTERNAL DATA if they also use Pine or IMAP. How can I fix this? question.

1.3 Can I set up a POP or IMAP server on Windows XP, 2000, NT, Me, 98, or 95?

Yes. Refer to the NT specific notes in files CONFIG and BUILD. Also, for DOS-based versions of Windows (Windows Me, 98, and 95) you *must* set up CRAM-MD5 authentication, as described in md5.txt.

There is no file access control on Windows 9x or Me, so you probably will have to do modifications to env_unix.c to prevent people from hacking others' mail.

Note, however, that the server is not plug and play the way it is for UNIX.

1.4 Can I set up a POP or IMAP server on Windows 3.1 or DOS?
1.5 Can I set up a POP or IMAP server on Macintosh?
1.6 Can I set up a POP or IMAP server on VAX/VMS?

Yes, it's just a small matter of programming.

1.7 Can I set up a POP or IMAP server on TOPS-20?

You have a TOPS-20 system? Cool.

If IMAP2 (RFC 1176) is good enough for you, you can use MAPSER which is about the ultimate gonzo pure TOPS-20 extended addressing assembly language program. Unfortunately, IMAP2 is barely good enough for Pine these days, and most other IMAP clients won't work with IMAP2 at all. Maybe someone will hack MAPSER to do IMAP4rev1 some day.

We don't know if anyone wrote a POP3 server for TOPS-20. There definitely was a POP2 server once upon a time.

Or you can port the POP and IMAP server from this IMAP toolkit to it. All that you need for a first stab is to port the MTX driver. That'll probably be just a couple of hours of hacking.

1.8 Are hierarchical mailboxes supported?
1.9 Are "dual-use" mailboxes supported?
1.10 Can I have a mailbox that has both messages and sub-mailboxes?

Yes. However, there is one important caveat.

Some mailbox formats, including the default which is the traditional UNIX mailbox format, are stored as a single file containing all the messages. UNIX does not permit a name in the filesystem to be both a file and a directory; consequently you can not have a sub-mailbox within a mailbox that is in one of these formats.

This is not a limitation of the software; this is a limitation of UNIX. For example, there are mailbox formats in which the name is a directory and each message is a file within that directory; these formats support sub-mailboxes within such mailboxes. However, for technical reasons, the "flat file" formats are generally preferred since they perform better. Read imap-2007/docs/formats.txt for more information on this topic.

It is always permissible to create a directory that is not a mailbox, and have sub-mailboxes under it. The easiest way to create a directory is to create a new mailbox inside a directory that doesn't already exist. For example, if you create "Mail/testbox" on UNIX, the directory "Mail/" will automatically be created and then the mailbox "testbox" will be created as a sub-mailbox of "Mail/".

It is also possible to create the name "Mail/" directly. Check the documentation for your client software to see how to do this with that software.

Of course, on Windows systems you would use "\" instead of "/".

1.11 What is the difference between "mailbox" and "folder"?

The term "mailbox" is IMAP-speak for what a lot of software calls a "folder" or a "mail folder". However, "folder" is often used in other contexts to refer to a directory, for example, in the graphic user interface on both Windows and Macintosh.

A "mailbox" is specifically defined as a named object that contains messages. It is not required to be capable of containing other types of objects including other mailboxes; although some mailbox formats will permit this.

In IMAP-speak, a mailbox which can not contain other mailboxes is called a "no-inferiors mailbox". Similarly, a directory which can not contain messages is not a mailbox and is called a "no-select name".

1.12 What is the status of internationalization?

The IMAP toolkit is partially internationalized and multilingualized.

Searching is supported in the following charsets: US-ASCII, UTF-8, ISO-8859-1, ISO-8859-2, ISO-8859-3, ISO-8859-4, ISO-8859-5, ISO-8859-6, ISO-8859-7, ISO-8859-8, ISO-8859-9, ISO-8859-10, ISO-8859-11, ISO-8859-13, ISO-8859-14, ISO-8859-15, ISO-8859-16, KOI8-R, KOI8-U (alias KOI8-RU), TIS-620, VISCII, ISO-2022-JP, ISO-2022-KR, ISO-2022-CN, ISO-2022-JP-1, ISO-2022-JP-2, GB2312 (alias CN-GB), CN-GB-12345, BIG5 (alias CN-BIG5), EUC-JP, EUC-KR, Shift_JIS, Shift-JIS, KS_C_5601-1987, KS_C_5601-1992, WINDOWS_874, WINDOWS-1250, WINDOWS-1251, WINDOWS-1252, WINDOWS-1253, WINDOWS-1254, WINDOWS-1255, WINDOWS-1256, WINDOWS-1257, WINDOWS-1258.

All ISO-2022-?? charsets are treated identically, and support ASCII, JIS Roman, hankaku katakana, ISO-8859-[1 - 10], TIS, GB 2312, JIS X 0208, JIS X 0212, KSC 5601, and planes 1 and 2 of CNS 11643.

EUC-JP includes support for JIS X 0212 and hankaku katakana.

c-client library support also exists to convert text in any of the above charsets into Unicode, including headers with MIME encoded-words.

There is no support for localization (e.g. non-English error messages) at the present time, but such support is planned.

1.13 Can I use SSL?

Yes. See the answer to the How do I configure SSL? question.

1.14 Can I use TLS and the STARTTLS facility?

Yes. See the answer to the How do I configure TLS and the STARTTLS facility? question.

1.15 Can I use CRAM-MD5 authentication?

Yes. See the answer to the How do I configure CRAM-MD5 authentication? question.

1.16 Can I use APOP authentication?

Yes. See the How do I configure APOP authentication? question.

Note that there is no client support for APOP authentication.

1.17 Can I use Kerberos V5?

Yes. See the answer to the How do I configure Kerberos V5? question.

1.18 Can I use PAM for plaintext passwords?

Yes. See the answer to the How do I configure PAM for plaintext passwords? question.

1.19 Can I use Kerberos 5 for plaintext passwords?

Yes. See the answer to the How do I configure Kerberos 5 for plaintext passwords? question.

1.20 Can I use AFS for plaintext passwords?

Yes. See the answer to the How do I configure AFS for plaintext passwords? question.

1.21 Can I use DCE for plaintext passwords?

Yes. See the answer to the How do I configure DCE for plaintext passwords? question.

1.22 Can I use the CRAM-MD5 database for plaintext passwords?

Yes. See the answer to the How do I configure the CRAM-MD5 database for plaintext passwords? question.

1.23 Can I disable plaintext passwords?

Yes. See the answer to the How do I disable plaintext passwords? question.

1.24 Can I disable plaintext passwords on unencrypted sessions, but allow them on encrypted sessions?

Yes. See the answer to the How do I disable plaintext passwords on unencrypted sessions, but allow them in SSL or TLS sessions? question.

1.25 Can I use virtual hosts?

Yes. See the answer to the How do I configure virtual hosts? question.

1.26 Can I use RPOP authentication?

There is no support for RPOP authentication.

1.27 Can I use Kerberos V4?

Kerberos V4 is not supported. Kerberos V4 client-only contributed code is available in

ftp://ftp.cac.washington.edu/mail/kerberos4-patches.tar.Z

This is a patchkit which must be applied to the IMAP toolkit according to the instructions in the patchkit's README. We can not promise that this code works.

1.28 Is there support for S/Key or OTP?

There is currently no support for S/Key or OTP. There may be an OTP SASL authenticator available from third parties.

1.29 Is there support for NTLM or SPA?

There is currently no support for NTLM or SPA, nor are there any plans to add such support. In general, I avoid vendor-specific mechanisms. I also believe that these mechanisms are being deprecated by their vendor.

There may be an NTLM SASL authenticator available from third parties.

1.30 Is there support for mh?

Yes, but only as a legacy format. Your mh format INBOX is accessed by the name "#mhinbox", and all other mh format mailboxes are accessed by prefixing "#mh/" to the name, e.g. "#mh/foo". The mh support uses the "Path:" entry in your .mh_profile file to identify the root directory of your mh format mailboxes.

Non-legacy use of mh format is not encouraged. There is no support for permanent flags or unique identifiers; furthermore there are known severe performance problems with the mh format.

1.31 Is there support for qmail and the maildir format?

There is no support for qmail or the maildir format in our distribution, nor are there any plans to add such support. Maildir support may be available from third parties.

1.32 Is there support for the Cyrus mailbox format?

No.

1.33 Is this software Y2K compliant?

Please read the files Y2K and calendar.txt.

2. What Do I Need to Build This Software?

2.1 What do I need to build this software with SSL on UNIX?

You need to build and install OpenSSL first.

2.2 What do I need to build this software with Kerberos V on UNIX?

You need to build and install MIT Kerberos first.

2.3 What do I need to use a C++ compiler with this software to build my own application?

If you are building an application using the c-client library, use the new c-client.h file instead of including the other include files. It seems that c-client.h should define away all the troublesome names that conflict with C++.

If you use gcc, you may need to use -fno-operator-names as well.

2.4 What do I need to build this software on Windows?

You need Microsoft Visual C++ 6.0, Visual C++ .NET, or Visual C# .NET (which you can buy from any computer store), along with the Microsoft Platform SDK (which you can download from Microsoft's web site).

You do not need to install the entire Platform SDK; it suffices to install just the Core SDK and the Internet Development SDK.

2.5 What do I need to build this software on DOS?

It's been several years since we last attempted to do this. At the time, we used Microsoft C.

2.6 Can't I use Borland C to build this software on the PC?

Probably not. If you know otherwise, please let us know.

2.7 What do I need to build this software on the Mac?

It has been several years since we last attempted to do this. At the time, we used Symantec THINK C; but today you'll need a C compiler which allows segments to be more than 32K.

2.8 What do I need to build this software on VMS?

You need the VMS C compiler, and either the Multinet or Netlib TCP.

2.9 What do I need to build this software on TOPS-20?

You need the TOPS-20 KCC compiler.

2.10 What do I need to build this software on Amiga or OS/2?

We don't know.

2.11 What do I need to build this software on Windows CE?

This port is incomplete. Someone needs to finish it.

3. Build and Configuration Questions

3.1 How do I configure the IMAP and POP servers on UNIX?
3.2 I built and installed the servers according to the BUILD instructions. It can't be that easy. Don't I need to write a config file?

For ordinary "vanilla" UNIX systems, this software is plug and play; just build it, install it, and you're done. If you have a modified system, then you may want to do additional work; most of this is to a single source code file (env_unix.c on UNIX systems). Read the file CONFIG for more details.

Yes, it's that easy. There are some additional options, such as SSL or Kerberos, which require additional steps to build. See the relevant questions below.

3.3 How do I make the IMAP and POP servers look for INBOX at some place other than the mail spool directory?
3.4 How do I make the IMAP server look for secondary folders at some place other than the user's home directory?

Please read the file CONFIG for discussion of this and other issues.

3.5 How do I configure SSL?
3.6 How do I configure TLS and the STARTTLS facility?

imap-2007 supports SSL and TLS client functionality on UNIX and 32-bit Windows for IMAP, POP3, SMTP, and NNTP; and SSL and TLS server functionality on UNIX for IMAP and POP3.

UNIX SSL build requires that a third-party software package, OpenSSL, be installed on the system first. Read imap-2007/docs/SSLBUILD for more information.

SSL is supported via undocumented Microsoft interfaces in Windows 9x and NT4; and via standard interfaces in Windows 2000, Windows Millenium, and Windows XP.

3.7 How do I build/install OpenSSL and obtain/create certificates for use with SSL?

If you need help in doing this, try the contacts mentioned in the OpenSSL README. We do not offer support for OpenSSL or certificates.

3.8 How do I configure CRAM-MD5 authentication?
3.9 How do I configure APOP authentication?

CRAM-MD5 authentication is enabled in the IMAP and POP3 client code on all platforms. Read md5.txt to learn how to set up CRAM-MD5 and APOP authentication on UNIX and NT servers.

There is no support for APOP client authentication.

3.10 How do I configure Kerberos V5?

imap-2007 supports client and server functionality on UNIX and 32-bit Windows.

Kerberos V5 is supported by default in Windows 2000 builds:

 nmake -f makefile.w2k

Other builds require that a third-party Kerberos package, e.g. MIT Kerberos, be installed on the system first.

To build with Kerberos V5 on UNIX, include EXTRAAUTHENTICATORS=gss in the make command line, e.g.

 make lnp EXTRAAUTHENTICATORS=gss

To build with Kerberos V5 on Windows 9x, Windows Millenium, and NT4, use the "makefile.ntk" file instead of "makefile.nt":


 nmake -f makefile.ntk

3.11 How do I configure PAM for plaintext passwords?

On Linux systems, use the lnp port, e.g.

 make lnp

On Solaris systems and other systems with defective PAM implementations, build with PASSWDTYPE=pmb, e.g.

 make sol PASSWDTYPE=pmb

On all other systems, build with PASSWDTYPE=pam, e.g

 make foo PASSWDTYPE=pam

If you build with PASSWDTYPE=pam and authentication does not work, try rebuilding (after a "make clean") with PASSWDTYPE=pmb.

3.12 It looks like all I have to do to make the server use Kerberos is to build with PAM on my Linux system, and set it up in PAM for Kerberos passwords. Right?

Yes and no.

Doing this will make plaintext password authentication use the Kerberos password instead of the /etc/passwd password.

However, this will NOT give you Kerberos-secure authentication. See the answer to the How do I configure Kerberos V5? question for how to build with Kerberos-secure authentication.

3.13 How do I configure Kerberos 5 for plaintext passwords?

Build with PASSWDTYPE=gss, e.g.

 make sol PASSWDTYPE=gss

However, this will NOT give you Kerberos-secure authentication. See the answer to the How do I configure Kerberos V5? question for how to build with Kerberos-secure authentication.

3.14 How do I configure AFS for plaintext passwords?

Build with PASSWDTYPE=afs, e.g

 make sol PASSWDTYPE=afs

3.15 How do I configure DCE for plaintext passwords?

Build with PASSWDTYPE=dce, e.g

 make sol PASSWDTYPE=dce

3.16 How do I configure the CRAM-MD5 database for plaintext passwords?

The CRAM-MD5 password database is automatically used for plaintext password if it exists.

Note that this is NOT CRAM-MD5-secure authentication. You probably want to consider disabling plaintext passwords for non-SSL/TLS sessions. See the next two questions.

3.17 How do I disable plaintext passwords?

Server-level plaintext passwords can be disabled by setting PASSWDTYPE=nul, e.g.

 make lnx EXTRAAUTHENTICATORS=gss PASSWDTYPE=nul

Note that you must have a CRAM-MD5 database installed or specify at least one EXTRAAUTHENTICATOR, otherwise it will not be possible to log in to the server.

When plaintext passwords are disabled, the IMAP server will advertise the LOGINDISABLED capability and the POP3 server will not advertise the USER capability.

3.18 How do I disable plaintext passwords on unencrypted sessions, but allow them in SSL or TLS sessions?

Do not set PASSWDTYPE=nul or SSLTYPE=unix. Set SSLTYPE=nopwd instead, e.g.

 make lnx SSLTYPE=nopwd

When plaintext passwords are disabled, the IMAP server will advertise the LOGINDISABLED capability and the POP3 server will not advertise the USER capability.

Plaintext passwords will always be enabled in SSL sessions; the IMAP server will not advertise the LOGINDISABLED capability and the POP3 server will advertise the USER capability.

If the client does a successful start-TLS in a non-SSL session, plaintext passwords will be enabled, and a new CAPABILITY or CAPA command (which is required after start-TLS) will show the effect as in SSL sessions.

3.19 How do I configure virtual hosts?

This is automatic, but with certain restrictions.

The most important one is that each virtual host must have its own IP address; otherwise the server has no way of knowing which virtual host is desired.

As distributed, the software uses a global password file; hence user "fred" on one virtual host is "fred" on all virtual hosts. You may want to modify the checkpw() routine to implement some other policy (e.g. separate password files).

Note that the security model assumes that all users have their own unique UNIX UID number. So if you use separate password files you should make certain that the UID numbers do not overlap between different files.

More advanced virtual host support may be available as patches from third parties.

3.20 Why do I get compiler warning messages such as:

 passing arg 3 of `scandir' from incompatible pointer type
 Pointers are not assignment-compatible.
 Argument #4 is not the correct type.

during the build?

You can safely ignore these messages.

Over the years, the prototype for scandir() has changed, and thus is variant across different UNIX platforms. In particular, the definitions of the third argument (type select_t) and fourth argument (type compar_t) have changed over the years, the issue being whether or not the arguments to the functions pointed to by these function pointers are of type const or not.

The way that c-client calls scandir() will tend to generate these compiler warnings on newer systems such as Linux; however, it will still build. The problem with fixing the call is that then it won't build on older systems.

3.21 Why do I get compiler warning messages such as

 Operation between types "void(*)(int)" and "void*" is not allowed.
 Function argument assignment between types "void*" and "void(*)(int)" is not allowed.
 Pointers are not assignment-compatible.
 Argument #5 is not the correct type.

during the build?

You can safely ignore these messages.

All known systems have no problem with casting a function pointer to/from a void* pointer, certain C compilers issue a compiler diagnostic because this facility is listed as a "Common extension" by the C standard:

 K.5.7  Function pointer casts
  [#1] A pointer to an object or to void may be cast to a pointer
       to a function, allowing data to be invoked as a function (6.3.4).
  [#2] A pointer to a function may be cast to a pointer to an
       object or to void, allowing a function to be inspected or
       modified (for example, by a debugger) (6.3.4).

It may be just a "common extension", but this facility is relied upon heavily by c-client.

3.22 Why do I get linker warning messages such as:

mtest.c:515: the `gets' function is dangerous and should not be used.

during the build? Isn't this a security bug?

You can safely ignore this message.

Certain linkers, most notably on Linux, give this warning message. It is indeed true that the traditional gets() function is not a safe one.

However, the mtest program is only a demonstration program, a model of a very basic application program using c-client. It is not something that you would install, much less run in any security-sensitive context.

mtest has numerous other shortcuts that you wouldn't want to do in a real application program.

The only "security bug" with mtest would be if it was run by some script in a security-sensitive context, but mtest isn't particularly useful for such purposes. If you wanted to write a script to automate some email task using c-client, you'd be better off using imapd instead of mtest.

mtest only has two legitimate uses. It's a useful testbed for me when debugging new versions of c-client, and it's useful as a model for someone writing a simple c-client application to see how the various calls work.

By the way, if you need a more advanced example of c-client programming than mtest (and you probably will), I recommend that you look at the source code for imapd and Pine.

3.23 Why do I get linker warning messages such as:

 auth_ssl.c:92: the `tmpnam' function is dangerous and should not be used.

during the build? Isn't this a security bug?

You can safely ignore this message.

Certain linkers, most notably on Linux, give this warning message, based upon two known issues with tmpnam():

there can be a buffer overflow if an inadequate buffer is allocated.

there can be a timing race caused by certain incautious usage of the return value.

Neither of these issues applies in the particular use that is made of tmpnam(). More importantly, the tmpnam() call is never executed on Linux systems.

3.24 OK, suppose I see a warning message about a function being "dangerous and should not be used" for something other than this gets() or tmpnam() call?

Please forward the details for investigation.

4. Operational Questions

4.1 How can I enable anonymous IMAP logins?

Create the file /etc/anonymous.newsgroups. At the present time, this file should be empty. This will permit IMAP logins as anonymous as well as the ANONYMOUS SASL authenticator. Anonymous users have access to mailboxes in the #news., #ftp/, and #public/ namespaces only.

4.2 How do I set up an alert message that each IMAP user will see?

Create the file /etc/imapd.alert with the text of the message. This text should be kept to one line if possible. Note that this will cause an alert to every IMAP user every time they initiate an IMAP session, so it should only be used for critical messages.

4.3 How does the c-client library choose which of its several mechanisms to use to establish an IMAP connection to the server? I noticed that it can connect on port 143, port 993, via rsh, and via ssh.

c-client chooses how to establish an IMAP connection via the following rules:

If /ssl is specified, use an SSL connection. Fail otherwise.
Else if client is a UNIX system and "ssh server exec /etc/rimapd" works, use that
Else if /tryssl is specified and an SSL connection works, use that.
Else if client is a UNIX system and "rsh server exec /etc/rimapd" works, use that.
Else use a non-SSL connection.

4.4 I am using a TLS-capable IMAP server, so I don't need to use /ssl to get encryption. However, I want to be certain that my session is TLS encrypted before I send my password. How to I do this?

Use the /tls option in the mailbox name. This will cause an error message and the connection to fail if the server does not negotiate STARTTLS.

4.5 How do I use one of the alternative formats described in the formats.txt document? In particular, I hear that mbx format will give me better performance and allow shared access.

The rumors about mbx format being preferred are true. It is faster than the traditional UNIX mailbox format and permits shared access.

However, and this is very important, note that using an alternative mailbox format is an advanced facility, and only expert users should undertake it. If you don't understand any of the following notes, you may not be enough of an expert yet, and are probably better off not going this route until you are more comfortable with your understanding.

Some of the formats, including mbx, are only supported by the software based on the c-client library, and are not recognized by other mailbox programs. The "vi" editor will corrupt any mbx format mailbox that it encounters.

Another problem is that the certain formats, including mbx, use advanced file access and locking techniques that do not work reliably with NFS. NFS is not a real filesystem. Use IMAP instead of NFS for distributed access.

Each of the following steps are in escalating order of involvement. The further you go down this list, the more deeply committed you become:

The simplest way to create a mbx-format mailbox is to prefix the name with "#driver.mbx/" when creating a mailbox through c-client. For example, if you create "#driver.mbx/foo", the mailbox "foo" will be created in mbx format. Only use "#driver.mbx/" when creating the mailbox. At all other times, just use the name ("foo" in this example); the software will automatically select the driver for mbx whenever that mailbox is accessed without you doing anything else.
You can use the "mailutil copy" command to copy an existing mailbox to a new mailbox in mbx format. Read the man page provided with the mailutil program for details.
If you create an mbx-format INBOX, by creating "#driver.mbx/INBOX" (note that "INBOX" must be all uppercase), then subsequent access to INBOX by any c-client based application will use the mbx-format INBOX. Any mail delivered to the traditional format mailbox in the spool directory (e.g. /var/spool/mail/$USER) will automatically be copied into the mbx-format INBOX and the spool directory copy removed.
You can cause any newly-created mailboxes to be in mbx-format by default by changing the definition of CREATEPROTO=unixproto to be CREATEPROTO=mbxproto in src/osdep/unix/Makefile, then rebuilding the IMAP toolkit (do a "make clean" first). Do not change EMPTYPROTO, since mbx format mailboxes are never a zero-byte file. If you use Pine or the imap-utils, you should probably also rebuild them with the new IMAP toolkit too.
You can deliver directly to the mbx-format INBOX by use of the tmail or dmail programs. tmail is for direct invocation from sendmail (or whatever MTA program you use); dmail is for calls from procmail. Both of these programs have man pages which must be read carefully before making this change.

Most other servers (e.g. Cyrus) require use of a non-standard format. A full-fledged format conversion is not significantly different from what you have to do with other servers. The difference, which makes format conversion procedures somewhat more complicated with this server, is that there is no "all or nothing" requirement with this server. There are many points in between. A format conversion can be anything from a single mailbox or single user, to systemwide.

This is good in that you can decide how far to go, or do the steps incrementally as you become more comfortable with the result. On the other hand, there's no "One True Way" which can be boiled down to a simple set of pedagogical instructions.

A number of sites have done full-fledged format conversions, and are reportedly quite happy with the results. Feel free to ask in the comp.mail.imap newsgroup or the imap-uw mailing list for advice or help.

4.6 How do I set up shared mailboxes?

At the simplest level, a shared mailbox is one which has UNIX file and directory protections which permit multiple users to access it. What this means is that your existing skills and tools to create and manage shared files on your UNIX system apply to shared mailboxes; e.g.

 chmod 666 mailbox

You may want to consider the use of a mailbox format which permits multiple simultaneous read/write sessions, such as the mbx format. The traditional UNIX format only allows one read/write session to a mailbox at a time.

An additional convenience item are three system directories, which can be set up for shared namespaces. These are: #ftp, #shared, and #public, and are defined by creating the associated UNIX users and home directories as described below.

#ftp/ refers to the anonymous ftp filesystem exported by the ftp server, and is equivalent to the home directory for UNIX user "ftp". For example, #ftp/foo/bar refers to the file /foo/bar in the anonymous FTP filesystem, or ~ftp/foo/bar for normal users. Anonymous FTP files are available to anonymous IMAP logins. By default, newly-created files in #ftp/ are protected 644.

#public/ refers to an IMAP toolkit convention called "public" files, and is equivalent to the home directory for UNIX user "imappublic". For example, #public/foo/bar refers to the file ~imappublic/foo/bar. Public files are available to anonymous IMAP logins. By default, newly-created files in #public are created with protection 0666.

#shared/ refers to an IMAP toolkit convention called "shared" files, and is equivalent to the home directory for UNIX user "imapshared". For example, #shared/foo/bar refers to the file ~imapshared/foo/bar. Shared files are not available to anonymous IMAP logins. By default, newly-created files in #shared are created with protection 0660.

4.7 How can I make the server syslogs go to someplace other than the mail syslog?

The openlog() call that sets the syslog facility is in src/osdep/unix/env_unix.c in routine server_init(). You need to edit this file to change the syslog facility from LOG_MAIL to the facility you want, then rebuild. You also need to set up your /etc/syslog.conf properly.

Refer to the man pages for syslog and syslogd for more information on what the available syslog facilities are and how to configure syslogs. If you still don't understand what to do, find a UNIX system expert.

5. Security Questions

5.1 I see that the IMAP server allows access to arbitary files on the system, including /etc/passwd! How do I disable this?

You should not worry about this if your IMAP users are allowed shell access. The IMAP server does not permit any access that the user can not have via the shell.

If, and only if, you deny your IMAP users shell access, you may want to consider one of three choices. Note that these choices reduce IMAP functionality, and may have undesirable side effects. Each of these choices involves an edit to file src/osdep/unix/env_unix.c

The first (and recommended) choice is to set restrictBox as described in file CONFIG. This will disable access to the filesystem root, to other users' home directory, and to superior directory.

The second (and strongly NOT recommended) choice is to set closedBox as described in file CONFIG. This puts each IMAP session into a so-called "chroot jail", and thus setting this option is extremely dangerous; it can make your system much less secure and open to root compromise attacks. So do not use this option unless you are absolutely certain that you understand all the issues of a "chroot jail."

The third choice is to rewrite routine mailboxfile() to implement whatever mapping from mailbox name to filesystem name (and restrictions) that you wish. This is the most general choice. As a guide, you can see at the start of routine mailboxfile() what the restrictBox choice does.

5.2 I've heard that IMAP servers are insecure. Is this true?

There are no known security problems in this version of the IMAP toolkit, including the IMAP and POP servers. The IMAP and POP servers limit what can be done while not logged in, and as part of the login process discard all privileges except those of the user.

As with other software packages, there have been buffer overflow vulnerabilities in past versions. All known problems of this nature are fixed in this version.

There is every reason to believe that the bad guys are engaged in an ongoing effort to find vulnerabilities in the IMAP toolkit. We look for such problems, and when one is found we fix it.

It's unfortunate that any vulnerabilities existed in past versions, and we're doing my best to keep the IMAP toolkit free of vulnerabilities. No new vulnerabilities have been discovered in quite a while, but efforts will not be relaxed.

Beware of vendors who claim that their implementations can not have vulnerabilities.

5.3 How do I know that I have the most secure version of the server?

The best way is to keep your server software up to date. The bad guys are always looking for ways to crack software, and when they find one, let all their friends know.

Oldtimers used to refer to a concept of software rot: if your software hasn't been updated in a while, it would "rot" -- tend to acquire problems that it didn't have when it was new.

The latest release version of the IMAP toolkit is always available at ftp://ftp.cac.washington.edu/mail/imap.tar.Z

5.4 I see all these strcpy() and sprintf() calls, those are unsafe, aren't they?

Yes and no.

It can be unsafe to do these calls if you do not know that the string being written will fit in the buffer. However, they are perfectly safe if you do know that.

Beware of programmers who advocate doing a brute-force change of all instances of

 strcpy (s,t);

 strncpy (s,t,n)[n] = '\0';

and similar measures in the name of "fixing all possible buffer overflows."

There are examples in which a security bug was introduced because of this type of "fix", due to the programmer using the wrong value for n. In one case, the programmer thought that n was larger than it actually was, causing a NUL to be written out of the buffer; in another, n was too small, and a security credential was truncated.

What is particularly ironic was that in both cases, the original strcpy() was safe, because the size of the source string was known to be safe.

With all this in mind, the software has been inspected, and it is believed that all places where buffer overflows can happen have been fixed. The strcpy()s that are still are in the code occur after a size check was done in some other way.

Note that the common C idiom of

 *s++ = c;

is just as vulnerable to buffer overflows. You can't cure buffer overflows by outlawing certain functions, nor is it desirable to do so; sometimes operations like strcpy() translate into fast machine instructions for better performance.

Nothing replaces careful study of code. That's how the bad guys find bugs. Security is not accomplished by means of brute-force shortcuts.

5.5 Those /tmp lock files are protected 666, is that really right?

Yes. Shared mailboxes won't work otherwise. Also, you get into accidental denial of service problems with old lock files left lying around; this happens fairly frequently.

The deliberate mischief that can be caused by fiddling with the lock files is small-scale; harassment level at most. There are many -- and much more effective -- other ways of harassing another user on UNIX. It's usually not difficult to determine the culprit.

Before worrying about deliberate mischief, worry first about things happening by accident!

6. Why Did You Do This Strange Thing? Questions

6.1 Why don't you use GNU autoconfig / automake / autoblurdybloop?

Autoconfig et al are not available on all the platforms where the IMAP toolkit is supported; and do not work correctly on some of the platforms where they do exist. Furthermore, these programs add another layer of complexity to an already complex process.

Coaxing software that uses autoconfig to build properly on platforms which were not specifically considered by that software wastes an inordinate amount of time. When (not if) autoconfig fails to do the right thing, the result is an inpenetrable morass to untangle in order to find the problem and fix it.

The concept behind autoconfig is good, but the execution is flawed. It rarely does the right thing on a platform that wasn't specifically considered. Human life is too short to debug autoconfig problems, especially since the current mechanism is so much easier.

6.2 Why do you insist upon a build with -g? Doesn't it waste disk and memory space?

From time to time a submitted port has snuck in without -g. This has always ended up causing problems. There are only two valid excuses for not using -g in a port:

The compiler does not support -g
An alternate form of -g is needed with optimization, e.g. -g3.

There will be no new ports added without -g (or a suitable alternative) being set.

-g has not been arbitrarily added to the ports which do not currently have it because we don't know if doing so would break the build. However, any support issues with one of those port will lead to the correct -g setting being determined and permanently added.

Processors are fast enough (and disk space is cheap enough) that -g should be automatic in all compilers with no way of turning it off, and /bin/strip should be a symlink to /bin/true. Human life is too short to deal with binaries built without -g. Such binaries should be a bad memory of the days of KIPS processors and disks that costs several dollars per kilobyte.

6.3 Why don't you make c-client a shared library?

All too often, shared libraries create far more problems than they solve.

Remember that you only gain the benefit of a shared library when there are multiple applications which use that shared library. Even without shared libraries, on most modern operating systems (and many ancient ones too!) applications will share their text segments between across multiple processes running the same application. This means that if your system only runs one application (e.g. imapd) that uses the c-client library, then you gain no benefit from making c-client a shared library even if it has 100 imapd processes. You will, however suffer added complexity.

If you have a server system that just runs imapd and ipop3d, then making c-client a shared library will save just one copy of c-client no matter how many IMAP/POP3 processes are running.

The problem with shared libraries is that you have to keep around a copy of the library every time something changes in the library that would affect the interface the library presents to the application. So, you end up having many copies of the same shared library.

If you don't keep multiple copies of the shared library, then one of two things happens. If there was proper versioning, then you'll get a message such as "cannot open shared object file" or "minor versions don't match" and the application won't run. Otherwise, the application will run, but will fail in mysterious ways.

Several sites and third-party distributors have modified the c-client makefile in order to make c-client be a shared library. When (not if) a c-client based application fails in mysterious ways because of a library compatibility problem, the result is a bug report. A lot of time and effort ends up getting wasted investigating such bug reports.

Memory is so cheap these days that it's not worth it. Human life is too short to deal with shared library compatibility problems.

6.4 Why don't you use iconv() for internationalization support?

iconv() is not ubiquitous enough.

6.5 Why is the IMAP server connected to the home directory by default?

The IMAP server has no way of knowing what you might call "mail" as opposed to "some other file"; in fact, you can use IMAP to access any file.

The IMAP server also doesn't know whether your preferred subdirectory for mailbox files is "mail/", ".mail/", "Mail/", "Mailboxes/", or any of a zillion other possibilities. If one such name were chosen, it would undoubtably anger the partisans of all the other names.

It is possible to modify the software so that the default connected directory is someplace else. Please read the file CONFIG for discussion of this and other issues.

6.6 I have a Windows system. Why isn't the server plug and play for me?

There is no standard for how mail is stored on Windows; nor a single standard SMTP server. The closest to either would be the SMTP server in Microsoft's IIS.

So there's no default by which to make assumptions. As the software is set up, it assumes that the each user has an Windows login account and private home directory, and that mail is stored on that home directory as files in one of the popular UNIX formats. It also assumes that there is some tool equivalent to inetd on UNIX that does the TCP/IP listening and server startup.

Basically, unless you're an email software hacker, you probably want to look elsewhere if you want IMAP/POP servers for Windows.

6.7 I looked at the UNIX SSL code and saw that you have the SSL data payload size set to 8192 bytes. SSL allows 16K; why aren't you using the full size?

This is to avoid an interoperability problem with:

PC IMAP clients that use Microsoft's SChannel.DLL (SSPI) for SSL support
Microsoft Exchange server (which also uses SChannel).

SChannel has a bug that makes it think that the maximum SSL data payload size is 16379 bytes -- 5 bytes too small. Thus, c-client has to make sure that it never transmits full sized SSL packets.

The reason for using 8K (as opposed to, say, 16379 bytes, or 15K, or...) is that it corresponds with the TCP buffer size that the software uses elsewhere for input; there's a slight performance benefit to having the two sizes correspond or at least be a multiple of each other. Also, it keeps the size as a power of two, which might be significant on some platforms.

There wasn't a significant difference that we could measure between 8K and 15K.

Microsoft has developed a hotfix for this bug. Look up MSKB article number 300562. Contrary to the article text which implies that this is a Pine issue, this bug also affects Microsoft Exchange server with any client that transmits full-sized SSL payloads.

6.8 Why is an mh format INBOX called #mhinbox instead of just INBOX?

It's a long story. In brief, the mh format driver is less functional than any of the other drivers. It turned out that there were some users (including high-level administrators) who tried mh years ago and no longer use it, but still had an mh profile left behind.

When the mh driver used INBOX, it would see the mh profile, and proceed to move the user's INBOX into the mh format INBOX. This caused considerable confusion as some things stopped working.

6.9 Why don't you support the maildir format?

It is technically difficult to support maildir in IMAP while maintaining acceptable performance, robustness, following the requirements of the IMAP protocol specification, and following the requirements of maildir.

No one has succeeded in accomplishing all four together. The various maildir drivers offered as patches all have these problems. The problem is exacerbated because this implementation supports multiple formats; consequently this implementation can't make any performance shortcuts by assuming that all the world is maildir.

We can't do a better job than the maildir fan community has done with their maildir drivers. Similarly, if the maildir fan community provides the maildir driver, they take on the responsibility for answering maildir-specific support questions. This is as it should be, and that is why maildir support is left to the maildir fan community.

6.10 Why don't you support the Cyrus format?

There's no point to doing so. An implementation which supports multiple formats will never do as well as one which is optimized to support one single format.

If you want to use Cyrus mailbox format, you should use the Cyrus server, which is the native implementation of that format and is specifically optimized for that format. That's also why Cyrus doesn't implement any other format.

6.11 Why is it creating extra forks on my SVR4 system?

This is because your system only has fcntl() style locking and not flock() style locking. fcntl() locking has a design flaw that causes a close() to release any locks made by that process on the file opened on that file descriptor, even if the lock was made on a different file descriptor.

This design flaw causes unexpected loss of lock, and consequent mailbox corruption. The workaround is to do certain "dangerous operations" in another fork, thus avoiding doing a close() in the vulnerable fork.

The best way to solve this problem is to upgrade your SVR4 (Solaris, AIX, HP-UX, SGI) or OSF/1 system to a more advanced operating system, such as Linux or BSD. These more advanced operating systems have fcntl() locking for compatibility with SVR4, but also have flock() locking.

Beware of certain SVR4 systems, such as AIX, which have an "flock()" function in their C library that is just a jacket that does an fcntl() lock. This is not a true flock(), and has the same design flaw as fcntl().

6.12 Why are you so fussy about the date/time format in the internal "From " line in traditional UNIX mailbox files? My other mail program just considers every line that starts with "From " to be the start of the message.

You just answered your own question. If any line that starts with "From " is treated as the start of a message, then every message text line which starts with "From " has to be quoted (typically by prefixing a ">" character). People complain about this -- "why did a > get stuck in my message?"

So, good mail reading software only considers a line to be a "From " line if it follows the actual specification for a "From " line. This means, among other things, that the day of week is fixed-format: "May 14", but "May 7" (note the extra space) as opposed to "May 7". ctime() format for the date is the most common, although POSIX also allows a numeric timezone after the year. For compatibility with ancient software, the seconds are optional, the timezone may appear before the year, the old 3-letter timezones are also permitted, and "remote from xxx" may appear after the whole thing.

Unfortunately, some software written by novices use other formats. The most common error is to have a variable-width day of month, perhaps in the erroneous belief that RFC 2822 (or RFC 822) defines the format of the date/time in the "From " line (it doesn't; no RFC describes internal formats). I've seen a few other goofs, such as a single-digit second, but these are less common.

If you are writing your own software that writes mailbox files, and you really aren't all that savvy with all the ins and outs and ancient history, you should seriously consider using the c-client library (e.g. routine mail_append()) instead of doing the file writes yourself. If you must do it yourself, use ctime(), as in:

 fprintf (mbx,"From %s@%h %s",user,host,ctime (time (0)));

rather than try to figure out a good format yourself. ctime() is the most traditional format and nobody will flame you for using it.

6.13 Why is traditional UNIX format the default format?

Compatibility with the past 30 or so years of UNIX history. This server is the only one that completely interoperates with legacy UNIX mail tools.

6.14 Why do you write this "DON'T DELETE THIS MESSAGE -- FOLDER INTERNAL DATA" message at the start of traditional UNIX and MMDF format mailboxes?

This pseudo-message serves two purposes.

First, it establishes the mailbox format even when the mailbox has no messages. Otherwise, a mailbox with no messages is a zero-byte file, which could be one of several formats.

Second, it holds mailbox metadata used by IMAP: the UID validity, the last assigned UID, and mailbox keywords. Without this metadata, which must be preserved even when the mailbox has no messages, the traditional UNIX format wouldn't be able to support the full capabilities of IMAP.

6.15 Why don't you stash the mailbox metadata in the first real message of the mailbox instead of writing this fake FOLDER INTERNAL DATA message?

In fact, that is what is done if the mailbox is non-empty and does not already have a FOLDER INTERNAL DATA message.

One problem with doing that is that if some external program removes the first message, the metadata is lost and must be recreated, thus losing any prior UID or keyword list status that IMAP clients may depend upon.

Another problem is that this doesn't help if the last message is deleted. This will result in an empty mailbox, and the necessity to create a FOLDER INTERNAL DATA message.

6.16 Why aren't "dual-use" mailboxes the default?

Compatibility with the past 30 or so years of UNIX history, not to mention compatibility with user expectations when using shell tools.

6.17 Why do you use ucbcc to build on Solaris?

It is a long, long story about why cc is set to ucbcc. You need to invoke the C compiler so that it links with the SVR4 libraries and not the BSD libraries, otherwise readdir() will return the wrong information.

Of all the names in the most common path, ucbcc is the only name to be found (on /usr/ccs/bin) that points to a suitable compiler. cc is likely to be /usr/ucb/cc which is absolutely not the compiler that you want. The real SVR4 cc is probably something like /opt/SUNWspro/bin/cc which is rarely in anyone's path by default.

ucbcc is probably a link to acc, e.g. /opt/SUNWspro/SC4.0/bin/acc, and is the UCB C compiler using the SVR4 libraries.

If ucbcc isn't on your system, then punt on the SUN C compiler and use gcc instead (the gso port instead of the sol port).

If, in spite of all the above warnings, you choose to change "ucbcc" to "cc", you will probably find that the -O2 needs to be changed to -O. If you don't get any error messages with -O2, that's a pretty good indicator that you goofed and are running the compiler that will link with the BSD libraries.

To recap:

The sol port is designed to be built using the UCB compiler using the SVR4 libraries. This compiler is "ucbcc", which is lunk to acc. You use -O2 as one of the CFLAGS.
If you build the sol port with the UCB compiler using the BSD libraries, you will get no error messages but you will get bad binaries (the most obvious symptom is dropping the first two characters return filenames from the imapd LIST command. This compiler also uses -O2, and is very often what the user gets from "cc". BEWARE
If you build the sol port with the real SVR4 compiler, which is often hidden away or unavailable on many systems, then you will get errors from -O2 and you need to change that to -O. But you will get a good binary. However, you should try it with -O2 first, to make sure that you got this compiler and not the UCB compiler using BSD libraries.

6.18 Why should I care about some old system with BSD libraries? cc is the right thing on my Solaris system!

Because there still are sites that use such systems. On those systems, the assumption that "cc" does the right thing will lead to corrupt binaries with no error message or other warning that anything is amiss.

Too many sites have fallen victim to this problem.

6.19 Why do you insist upon writing .lock files in the spool directory?

Compatibility with the past 30 years of UNIX software which deals with the spool directory, especially software which delivers mail. Otherwise, it is possible to lose mail.

6.20 Why should I care about compatibility with the past?

This is one of those questions in which the answer never convinces those who ask it. Somehow, everybody who ever asks this question ends up answering it for themselves as they get older, with the very answer that they rejected years earlier.

7. Problems and Annoyances

7.1 Help! My INBOX is empty! What happened to my messages?

If you are seeing "0 messages" when you open INBOX and you know you have messages there (and perhaps have looked at your mail spool file and see that messages are there), then probably there is something wrong with the very first line of your mail spool file. Make sure that the first five bytes of the file are "From ", followed by an email address and a date/time in ctime() format, e.g.:

 From fred@foo.bar Mon May  7 20:54:30 2001

7.2 Help! All my messages in a non-INBOX mailbox have been concatenated into one message which claims to be from me and has a subject of the file name of the mailbox! What's going on?

Something wrong with the very first line of the mailbox. Make sure that the first five bytes of the file are "From ", followed by an email address and a date/time in ctime() format, e.g.:

 From fred@foo.bar Mon May  7 20:54:30 2001

7.3 Why do I get the message: CREATE failed: Can't create mailbox node xxxxxxxxx: File exists and how do I fix it?

See the answer to the Are hierarchical mailboxes supported? question.

7.4 Why can't I log in to the server? The user name and password are right!

There are a myriad number of possible answers to this question. The only way to say for sure what is wrong is run the server under a debugger such as gdb while root (yes, you must be root) with a breakpoint at routines checkpw() and loginpw(), then single-step until you see which test rejected you. The server isn't going to give any error messages other than "login failed" in the name of not giving out any unnecessary information to unauthorized individuals.

Here are some of the more common reasons why login may fail:

You didn't really give the correct user name and/or password.
Your client doesn't send the LOGIN command correctly; for example, IMAP2 clients won't send a password containing a "*" correctly to an IMAP4 server.
If you have set up a CRAM-MD5 database, remember that the password used is the one in the CRAM-MD5 database, and furthermore that there must also be an entry in /etc/passwd (but the /etc/passwd password is not used).
If you are using PAM, have you created a service file for the server in /etc/pam.d?
If you are using shadow passwords, have you used an appropriate port when building? In particular, note that "lnx" is for Linux systems without shadow passwords; you probably want "slx" or "lnp" instead.
If your system has account or password expirations, check to see that the expiration date hasn't passed.
You can't log in as root or any other UID 0 user. This is for your own safety, not to mention the fact that the servers use UID 0 as meaning "not logged in".

7.5 Help! My load average is soaring and I see hundreds of POP and IMAP servers, many logged in as the same user!

Certain inferior losing GUI mail reading programs have a "synchronize all mailboxes at startup" (IMAP) or "check for new mail every second" (POP) feature which causes a rapid and unchecked spawning of servers.

This is not a problem in the server; the client is really asking for all those server sessions. Unfortunately, there isn't much that the POP and IMAP servers can do about it; they don't spawned themselves.

Some sites have added code to record the number of server sessions spawned per user per hour, and disable login for a user who has exceeded a predetermined rate. This doesn't stop the servers from being spawned; it just means that a server session will commit suicide a bit faster.

Another possibility is to detect excessive server spawning activity at the level where the server is spawned, which would be inetd or possibly tcpd. The problem here is that this is a hard time to quantify. 50 sessions in a minute from a multi-user timesharing system may be perfectly alright, whereas 10 sessions a minute from a PC may be too much.

The real solution is to fix the client configuration, by disabling those evil features. Also tell the vendors of those clients how you feel about distributing denial-of-service attack tools in the guise of mail reading programs.

7.6 Why does mail disappear even though I set "keep mail on server"?
7.7 Why do I get the message Moved ##### bytes of new mail to /home/user/mbox from /var/spool/mail/user and why did this happen?

This is probably caused by the mbox driver. If the file "mbox" exists on the user's home directory and is in UNIX mailbox format, then when INBOX is opened this file will be selected as INBOX instead of the mail spool file. Messages will be automatically transferred from the mail spool file into the mbox file.

To disable this behavior, delete "mbox" from the EXTRADRIVERS list in the top-level Makefile and rebuild. Note that if you do this, users won't be able to access the messages that have already been moved to mbox unless they open mbox instead of INBOX.

7.8 Why isn't it showing the local host name as a fully-qualified domain name?
7.9 Why is the local host name in the From/Sender/Message-ID headers of outgoing mail not coming out as a fully-qualified domain name?

Your UNIX system is misconfigured. The entry for your system in /etc/hosts must have the fully-qualified domain name first, e.g.

 105.69.1.234	myserver.example.com myserver

A common mistake of novice system administrators is to have the short name first, e.g.

 105.69.1.234	myserver myserver.example.com

or to omit the fully qualified domain name entirely, e.g.

 105.69.1.234	myserver

The result of this is that when the IMAP toolkit does a gethostbyname() call to get the fully-qualified domain name, it would get "myserver" instead of "myserver.example.com".

On some systems, a configuration file (typically named /etc/svc.conf, /etc/netsvc.conf, or /etc/nsswitch.conf) can be used to configure the system to use the domain name system (DNS) instead of /etc/hosts, so it doesn't matter if /etc/hosts is misconfigured.

Check the man pages for gethostbyname, hosts, svc, and/or netsvc for more information.

Unfortunately, certain vendors, most notably SUN, have failed to make this clear in their documentation. Most of SUN's documentation assumes a corporate network that is not connected to the Internet.

net.folklore once (late 1980s) held that the proper procedure was to append the results of getdomainname() to the name returned by gethostname(), and some versions of sendmail configuration files were distributed that did this. This was incorrect; the string returned from getdomainname() is the Yellow Pages (a.k.a NIS) domain name, which is a completely different (albeit unfortunately named) entity from an Internet domain. These were often fortuitously the same string, except when they weren't. Frequently, this would result in host names with spuriously doubled domain names, e.g.

 myserver.example.com.example.com

This practice has been thoroughly discredited for many years, but folklore dies hard.

7.10 What does the message: Mailbox vulnerable - directory /var/spool/mail must have 1777 protection mean? How can I fix this?

In order to update a mailbox in the default UNIX format, it is necessary to create a lock file to prevent the mailer from delivering mail while an update is in progress. Some systems use a directory protection of 775, requiring that all mail handling programs be setgid mail; or of 755, requiring that all mail handling programs be setuid root.

The IMAP toolkit does not run with any special privileges, and I plan to keep it that way. It is antithetical to the concept of a toolkit if users can't write their own programs to use it. Also, I've had enough bad experiences with security bugs while running privileged; the IMAP and POP servers have to be root when not logged in, in order to be able to log themselves in. I don't want to go any deeper down that slippery slope.

Directory protection 1777 is secure enough on most well-managed systems. If you can't trust your users with a 1777 mail spool (petty harassment is about the limit of the abuse exposure), then you have much worse problems then that.

If you absolutely insist upon requiring privileges to create a lock file, external file locking can be done via a setgid mail program named /etc/mlock (this is defined by LOCKPGM in the c-client Makefile). If the toolkit is unable to create a <...mailbox...>.lock file in the directory by itself, it will try to call mlock to do it. I do not recommend doing this for performance reasons.

A sample mlock program is included as part of imap-2007. We have tried to make this sample program secure, but it has not been thoroughly audited.

7.11 What does the message: Mailbox is open by another process, access is readonly mean? How do I fix this?

A problem occurred in applying a lock to a /tmp lock file. Either some other program has the mailbox open and won't relenquish it, or something is wrong with the protection of /tmp or the lock.

Make sure that the /tmp directory is protected 1777. Some security scripts incorrectly set the protection of the /tmp directory to 775, which disables /tmp for all non-privileged programs.

7.12 What does the message: Can't get write access to mailbox, access is readonly mean?

The mailbox file is write-protected against you.

7.13 I set my POP3 client to "delete messages from server" but they never get deleted. What is wrong?

Make sure that your mailbox is not read-only: that the mailbox is owned by you and write enabled (protection 0600), and that the /tmp directory is longer world-writeable. /tmp must be world-writeable because lots of applications use it for scratch space. To fix this, do


 chmod 1777 /tmp

as root.

Make sure that your POP3 client issues a QUIT command when it finishes. The POP3 protocol specifies that deletions are discarded unless a proper QUIT is done.

Make sure that you are not opening multiple POP3 sessions to the same mailbox. It is a requirement of the POP3 protocol than only one POP3 session be in effect to a mailbox at a time, however some, poorly-written POP3 clients violate this. Also, some background "check for new mail" tasks also cause a violation. See the answer to the What does the syslog message: Killed (lost mailbox lock) user=... host=... mean? question for more details.

7.14 What do messages such as:

 Message ... UID ... already has UID ...
 Message ... UID ... less than ...
 Message ... UID ... greater than last ...
 Invalid UID ... in message ..., rebuilding UIDs

mean?

Something happened to corrupt the unique identifier regime in the mailbox. In traditional UNIX-format mailboxes, this can happen if the user deleted the "DO NOT DELETE" internal message.

This problem is relatively harmless; a new valid unique identifier regime will be created. The main effect is that any references to the old UIDs will no longer be useful.

So, unless it is a chronic problem or you feel like debugging, you can safely ignore these messages.

7.15 What do the error messages:

 Unable to read internal header at ...
 Unable to find CRLF at ...
 Unable to parse internal header at ...
 Unable to parse message date at ...
 Unable to parse message flags at ...
 Unable to parse message UID at ...
 Unable to parse message size at ...
 Last message (at ... ) runs past end of file ...

mean? I am using mbx format.

The mbx-format mailbox is corrupted and needs to be repaired.

You should make an effort to find out why the corruption happened. Was there an obvious system problem (crash or disk failure)? Did the user accidentally access the file via NFS? Mailboxes don't get corrupted by themselves; something caused the problem.

Some people have developed automated scripts, but if you're comfortable using emacs it's pretty easy to fix it manually. Do not use vi or any other editor unless you are certain that editor can handle binary!!!

If you are not comfortable with emacs, or if the file is too large to read with emacs, see the "step-by-step" technique later on for another way of doing it.

After the word "at" in the error message is the byte position it got to when it got unhappy with the file, e.g. if you see:

 Unable to parse internal header at 43921: ne bombastic blurdybloop

The problem occurs at the 43,931 byte in the file. That's the point you need to fix. c-client is expecting an internal header at that byte number, looking something like:

 6-Jan-1998 17:42:24 -0800,1045;000000100001-00000001

The format of this internal line is:

 dd-mmm-yyyy hh:mm:ss +zzzz,ssss;ffffffffFFFF-UUUUUUUU

The only thing that is variable is the "ssss" field, it can be as many digits as needed. All other fields (inluding the "dd") are fixed width. So, the easiest thing to do is to look forward in the file for the next internal header, and delete everything from the error point to that internal header.

Here's what to do if you want to be smarter and do a little bit more work. Generally, you're in the middle of a message, and there's nothing wrong with that message. The problem happened in the *previous* message. So, search back to the previous internal header. Now, remember that "ssss" field? That's the size of that message.

Mark where you are in the file, move the cursor to the line after the internal header, and skip that many bytes ("ssss") forward. If you're at the point of the error in the file, then that message is corrupt. If you're at a different point, then perhaps the previous message is corrupt and has a too long size count that "ate" into this message.

Basically, what you need to do is make sure that all those size counts are right, and that moving "ssss" bytes from the line after the internal header will land you at another internal header.

Usually, once you know what you're looking at, it's pretty easy to work out the corruption, and the best remedial action. Repair scripts will make the problem go away but may not always do the smartest/best salvage of the user's data. Manual repair is more flexible and usually preferable.

Here is a step-by-step technique for fixing corrupt mbx files that's a bit cruder than the procedure outlined above, but works for any size file.

In this example, suppose that the corrupt file is INBOX, the error message is

 Unable to find CRLF at 132551754

and the size of the INBOX file is 132867870 bytes.

The first step is to split the mailbox file at the point of the error:

Rename the INBOX file to some other name, such as INBOX.bad.
Copy the first 132,551,754 bytes of INBOX.bad to another file, such as INBOX.new.
Extract the trailing 316,116 bytes (132867870-132551754) of INBOX.bad into another file, such as INBOX.tail.
You no longer need INBOX.bad. Delete it.

In other words, use the number from the "Unable to find CRLF at" as the point to split INBOX into two new files, INBOX.new and INBOX.tail.

Now, remove the erroneous data:

Verify that you can open INBOX.new in IMAP or Pine.
The last message of INBOX.new is probably corrupted. Copy it to another file, such as badmsg.1, then delete and expunge that last message from INBOX.new
Locate the first occurance of text in INBOX.tail which looks like an internal header, as described above.
Remove all the text which occurs prior to that point, and place it into another file, such as badmsg.2. Note that in the case of a single digit date, there is a leading space which must not be removed (e.g. " 6-Nov-2001" not "6-Nov-2001").

Reassemble the mailbox:

Append INBOX.tail to INBOX.new.
You no longer need INBOX.tail. Delete it.
Verify that you can open INBOX.new in IMAP or Pine.

Reinstall INBOX.new as INBOX:

Check to see if you have received any new messages while repairing INBOX.
If you haven't received any new messages while repairing INBOX, just rename INBOX.new to INBOX.
If you have received new messages, be sure to copy the new messages from INBOX to INBOX.new before doing the rename.

You now have a working INBOX, as well as two files with corrupted data (badmsg.1 and badmsg.2). There may be some useful data in the two badmsg files that you might want to try salvaging; otherwise you can delete the two badmsg files.

7.16 What do the syslog messages:


 imap/tcp server failing (looping)
 pop3/tcp server failing (looping)

mean? When it happens, the listed service shuts down. How can I fix this?

The error message "server failing (looping), service terminated" is not from either the IMAP or POP servers. Instead, it comes from inetd, the daemon which listens for TCP connections to a number of servers, including the IMAP and POP servers.

inetd has a limit of 40 new server sessions per minute for any particular service. If more than 40 sessions are initiated in a minute, inetd will issue the "failing (looping), service terminated" message and shut down the service for 10 minutes. inetd does this to prevent system resource consumption by a client which is spawning infinite numbers of servers. It should be noted that this is a denial of service; however for some systems the alternative is a crash which would be a worse denial of service!

For larger server systems, the limit of 40 is much too low. The limit was established many years ago when a system typically only ran a few dozen servers.

On some versions of inetd, such as the one distributed with most versions of Linux, you can modify the /etc/inetd.conf file to have a larger number of servers by appending a period followed by a number after the nowait word for the server entry. For example, if your existing /etc/inetd.conf line reads:

 imap    stream  tcp     nowait  root    /usr/etc/imapd imapd

try changing it to be:

 imap    stream  tcp     nowait.100  root    /usr/etc/imapd imapd

Another example (using TCP wrappers):

 imap    stream  tcp     nowait  root    /usr/sbin/tcpd  imapd

try changing it to be:

 imap    stream  tcp     nowait.100  root    /usr/sbin/tcpd  imapd

to increase the limit to 100 sessions/minute.

Before making this change, please read the information in "man inetd" to determine whether or not your inetd has this feature. If it does not, and you make this change, the likely outcome is that you will disable IMAP service entirely.

Another way to fix this problem is to edit the inetd.c source code (provided by your UNIX system vendor) to set higher limits, rebuild inetd, install the new binary, and reboot your system. This should only be done by a UNIX system expert. In the inetd.c source code, the limits TOOMANY (normally 40) is the maximum number of new server sessions permitted per minute, and RETRYTIME (normally 600) is the number of seconds inetd will shut down the server after it exceeds TOOMANY.

7.17 What does the syslog message: Mailbox lock file /tmp/.600.1df3 open failure: Permission denied mean?

This usually means that some "helpful" security script person has protected /tmp so that it is no longer world-writeable. /tmp must be world-writeable because lots of applications use it for scratch space. To fix this, do

 chmod 1777 /tmp

as root.

If that isn't the answer, check the protection of the named file. If it is something other than 666, then either someone is hacking or some "helpful" person modified the code to have a different default lock file protection.

7.18 What do the syslog messages:

 Command stream end of file, while reading line user=... host=...
 Command stream end of file, while reading char user=... host=...
 Command stream end of file, while writing text user=... host=...

mean?

This message occurs when the session is disconnected without a proper LOGOUT (IMAP) or QUIT (POP) command being received by the server first.

In many cases, this is perfectly normal; many client implementations are impolite and do this. Some programmers think this sort of rudeness is "more efficient".

The condition could, however, indicate a client or network connectivity problem. The server has no way of knowing whether there's a problem or just a rude client, so it issues this message instead of a Logout.

Certain inferior losing clients disconnect abruptly after a failed login, and instead of saying that the login failed, just say that they can't access the mailbox. They then complain to the system manager, who looks in the syslog and finds this message. Not very helpful, eh? See the answer to the Why can't I log in to the server? The user name and password are right! question.

If the user isn't reporting a problem, you can probably ignore this message.

7.19 Why did my POP or IMAP session suddenly disconnect? The syslog has the message: Killed (lost mailbox lock) user=... host=...

This message only happens when either the traditional UNIX mailbox format or MMDF format is in use. This format only allows one session to have the mailbox open read/write at a time.

The servers assume that if a second session attempts to open the mailbox, that means that the first session is probably owned by an abandoned client. The common scenario here is a user who leaves his client running at the office, and then tries to read his mail from home. Through an internal mechanism called kiss of death, the second session requests the first session to kill itself. When the first session receives the "kiss of death", it issues the "Killed (lost mailbox lock)" syslog message and terminates. The second session then seizes read/write access, and becomes the new "first" session.

Certain poorly-designed clients routinely open multiple sessions to the same mailbox; the users of those clients tend to get this message a lot.

Another cause of this message is a background "check for new mail" task which does its work by opening a POP session to server every few seconds. They do this because POP doesn't have a way to announce new mail.

The solution to both situations is to replace the client with a good online IMAP client such as Pine. Life is too short to waste on POP clients and poorly-designed IMAP clients.

7.20 Why does my IMAP client show all the files on the system, recursively from the UNIX root directory?
7.21 Why does my IMAP client show all of my files, recursively from my UNIX home directory?

A well-written client should only show one level of hierarchy and then stop, awaiting explicit user action before going lower. However, some poorly-designed clients will recursively list all files, which may be a very long list (especially if you have symbolic links to directories that create a loop in the filesystem graph!).

This behavior has also been observed in some third-party c-client drivers, including maildir drivers. Consequently, this problem has even been observed in Pine. It is important to understand that this is not a problem in Pine or c-client; it is a problem in the third-party driver. A Pine built without that third-party driver will not have this problem.

7.22 Why does my IMAP client show that I have mailboxes named "#mhinbox", "#mh", "#shared", "#ftp", "#news", and "#public"?

These are IMAP namespace names. They represent other hierarchies in which messages may exist. These hierarchies may not necessarily exist on a server, but the namespace name is still in the namespace list in order to mark it as reserved.

A few poorly-designed clients display all namespace names as if they were top-level mailboxes in a user's list of mailboxes, whether or not they actually exist. This is a flaw in those clients.

7.23 Why does my IMAP client show all my files in my home directory?

As distributed, the IMAP server is connected to your home directory by default. It has no way of knowing what you might call "mail" as opposed to "some other file"; in fact, you can use IMAP to access any file.

Most clients have an option to configure your connected directory on the IMAP server. For example, in Pine you can specify this as the "Path" in your folder-collection, e.g.

 Nickname  : Secondary Folders
 Server    : imap.example.com
 Path      : mail/
 View      :

In this example, the user is connected to the "mail" subdirectory of his home directory.

Other servers call this the "folder prefix" or similar term.

It is possible to modify the IMAP server so that all users are automatically connected to some other directory, e.g. a subdirectory of the user's home directory. Read the file CONFIG for more details.

7.24 Why is there a long delay before I get connected to the IMAP or POP server, no matter what client I use?

There are two common occurances of this problem:

You are running a system (e.g. certain versions of Linux) which by default attempts to connect to an "IDENT" protocol (port 113) server on your client. However, a firewall or NAT box is blocking connections to that port, so the connection attempt times out.
The IDENT protocol is a well-known bad idea that does not deliver any real security but causes incredible problems. The idea is that this will give the server a record of the user name, or at least what some program listening on port 113 says is the user name. So, if somebody coming from port nnnnn on a system does something bad, IDENT may give you the userid of the bad guy.

The problem is, IDENT is only meaningful on a timesharing system which has an administrator who is privileged and users who are not. It is of no value on a personal system which has no separate concept of "system administrator" vs. "unprivileged user".

On either type of system, security-minded people either turn IDENT off or replace it with an IDENT server that lies. Among other things, IDENT gives spammers the ability to harvest email addresses from anyone who connects to a web page.

This problem has been showing up quite frequently on systems which use xinetd instead of inetd. Look for files named /etc/xinetd.conf, /etc/xinetd.d/imapd, /etc/inetd.d/ipop2d, and /etc/xinetd.d/ipop3d. In those files, look for lines containing "USERID", e.g.
```
 log_on_success += USERID
```
Hunt down such lines, and delete them ruthlessly from all files in which they occur. Don't be shy about it.
The DNS is taking a long time to do a reverse DNS (PTR record) lookup of the IP address of your client. This is a problem in your DNS, which either you or you ISP need to resolve. Ideally, the DNS should return the client's name; but if it can't it should at least return an error quickly.

As you may have noticed, neither of these are actual problems in the IMAP or POP servers; they are configuration issues with either your system or your network infrastructure. If this is all new to you, run (don't walk) to the nearest technical bookstore and get yourself a good pedagogical text on system administration for the type of system you are running.

7.25 Why is there a long delay in Pine or any other c-client based application call before I get connected to the IMAP server? The hang seems to be in the c-client mail_open() call. I don't have this problem with any other IMAP client. There is no delay connecting to a POP3 or NNTP server with mail_open().

By default, the c-client library attempts to make a connection through rsh (and ssh, if you enable that). If the command:

 rsh imapserver exec /etc/rimapd

(or ssh if that is enabled) returns with a "* PREAUTH" response, it will use the resulting rsh session as the IMAP session and not require an authentication step on the server.

Unfortunately, rsh has a design error that treats "TCP connection refused" as "temporary failure, try again"; it expects the "rsh not allowed" case to be implemented as a successful connection followed by an error message and close the connection.

It must be emphasized that this is a bug in rsh. It is not a bug in the IMAP toolkit.

The use of rsh can be disabled in any the following ways:

You can disable it for this particular session by either:
- setting an explicit port number in the mailbox name, e.g.
```
 {imapserver.foo.com:143}INBOX
```
- using SSL (the /ssl switch)
You can disable rsh globally by setting the rsh timeout value to 0 with the call:
```
 mail_parameters (NIL,SET_RSHTIMEOUT,0);
```

7.26 Why does a message sometimes get split into two or more messages on my SUN system?

This is caused by an interaction of two independent design problems in SUN mail software. The first problem is that the "forward message" option in SUN's mail tool program includes the internal "From " header line in the text that it forwarded. This internal header line is specific to traditional UNIX mailbox files and is not suitable for use in forwarded messages.

The second problem is that the mail delivery agent assumes that mail reading programs will not use the traditional UNIX mailbox format but instead an incompatible variant that depends upon a Content-Length: message header. Content-Length is widely recognized to have been a terrible mistake, and is no longer recommended for use in mail (it is used in other facilities that use MIME).

One symptom of the problem is that under certain circumstances, a message may get broken up into several messages. I'm also aware of security bugs caused by programs that foolishly trust "Content-Length:" headers with evil values.

To fix the mailer on your system, edit your sendmail.cf to change the Mlocal line to have the -E flag. A typical entry will lool like:

 Mlocal,	P=/usr/lib/mail.local, F=flsSDFMmnPE, S=10, R=20,
		A=mail.local -d $u

This fix will also work around the problem with mail tool, because it will insert a ">" before the internal header line to prevent it from being interpreted by mail reading software as an internal header line.

7.27 Why did my POP or IMAP session suddenly disconnect? The syslog has the message:

 Autologout user=<...my user name...> host=<...my client system...>

This is a problem in your client.

In the case of IMAP, it failed to communicate with the IMAP server for over 30 minutes; in the case of POP, it failed to communicate with the POP server for over 10 minutes.

7.28 What does the UNIX error message: TLS/SSL failure: myserver: SSL negotiation failed mean?
7.29 What does the PC error message: TLS/SSL failure: myserver: Unexpected TCP input disconnect mean?

This usually means that an attempt to negotiate TLS encryption via the STARTTLS command failed, because the server advertises STARTTLS functionality, but doesn't actually have it (e.g. because no certificates are installed).

Use the /notls option in the mailbox name to disable TLS negotiation.

7.30 What does the error message: TLS/SSL failure: myserver: Server name does not match certificate mean?

An SSL or TLS session encryption failed because the server name in the server's certificate does not match the name that you gave it. This could indicate that the server is not really the system you think that it is, but can be also be called if you gave a nickname for the server or name that was not fully-qualified. You must use the fully-qualified domain name for the server in order to validate its certificate

Use the /novalidate-cert option in the mailbox name to disable validation of the certificate.

7.31 What does the UNIX error message: TLS/SSL failure: myserver: self-signed certificate mean?
7.32 What does the PC error message: TLS/SSL failure: myserver: Self-signed certificate or untrusted authority mean?

An SSL or TLS session encryption failed because your server's certificate is "self-signed"; that is, it is not signed by any Certificate Authority (CA) and thus can not be validated. A CA-signed certificate costs money, and some smaller sites either don't want to pay for it or haven't gotten one yet. The bad part about this is that this means there is no guarantee that the server is really the system you think that it is.

Use the /novalidate-cert option in the mailbox name to disable validation of the certificate.

7.33 What does the UNIX error message: TLS/SSL failure: myserver: unable to get local issuer certificate mean?

An SSL or TLS session encryption failed because your system does not have the Certificate Authority (CA) certificates installed on OpenSSL's certificates directory. On most systems, this directory is /usr/local/ssl/certs). As a result, it is not possible to validate the server's certificate.

If CA certificates are properly installed, you should see factory.pem and about a dozen other .pem names such as thawteCb.pem.

As a workaround, you can use the /novalidate-cert option in the mailbox name to disable validation of the certificate; however, note that you are then vulnerable to various security attacks by bad guys.

The correct fix is to copy all the files from the certs/ directory in the OpenSSL distribution to the /usr/local/ssl/certs (or whatever) directory. Note that you need to do this after building OpenSSL, because the OpenSSL build creates a number of needed symbolic links. For some bizarre reason, the OpenSSL "make install" doesn't do this for you, so you must do it manually.

7.34 Why does reading certain messages hang when using Netscape? It works fine with Pine!

There are two possible causes.

Check the mail syslog. If you see the message "Killed (lost mailbox lock)" for the impacted user(s), read the FAQ entry regarding that message.

Check the affected mailbox to see if there are embedded NUL characters in the message. NULs in message texts are a technical violation of both the message format and IMAP specifications. Most clients don't care, but apparently Netscape does.

You can work around this by rebuilding imapd with the NETSCAPE_BRAIN_DAMAGE option set (see src/imapd/Makefile); this will cause imapd to convert all NULs to 0x80 characters. A better solution is to enable the feature in your MTA to MIME-convert messages with binary content. See the documentation for your MTA for how to do this.

7.35 Why does Netscape say that there's a problem with the IMAP server and that I should "Contact your mail server administrator."?

Certain versions of Netscape do this when you click the Manage Mail button, which uses an undocumented feature of Netscape's proprietary IMAP server.

You can work around this by rebuilding imapd with the NETSCAPE_BRAIN_DAMAGE option set (see src/imapd/Makefile) to a URL that points either to an alternative IMAP client (e.g. Pine) or perhaps to a homebrew mail account management page.

7.36 Why is one user creating huge numbers of IMAP or POP server sessions?

The user is probably using Outlook Express, Eudora, or a similar program. See the answer to the Help! My load average is soaring and I see hundreds of POP and IMAP servers, many logged in as the same user! question.

7.37 Why don't I get any new mail notifications from Outlook Express or Outlook after a while?

This is a known bug in Outlook Express. Microsoft is aware of the problem and its cause. They have informed us that they do not have any plans to fix it at the present time.

The problem is also reported in Outlook 2000, but not verified.

Outlook Express uses the IMAP IDLE command to avoid having to "ping" the server every few minutes for new mail. Unfortunately, Outlook Express overlooks the part in the IDLE specification which requires that a client terminate and restart the IDLE before the IMAP 30 minute inactivity autologout timer triggers.

When this happens, Outlook Express displays "Not connected" at the bottom of the window. Since it's no longer connected to the IMAP server, it isn't going to notice any new mail.

As soon as the user does anything that would cause an IMAP operation, Outlook Express will reconnect and new mail will flow again. If the user does something that causes an IMAP operation at least every 29 minutes, the problem won't happen.

Modern versions of imapd attempt to work around the problem by automatically reporting fake new mail after 29 minutes. This causes Outlook Express to exit the IDLE state; as soon as this happens imapd revokes the fake new mail. As long as this behavior isn't known to cause problems with other clients, this workaround will remain in imapd.

7.38 Why don't I get any new mail notifications from Entourage?

This is a known bug in Entourage.

You built an older version of imapd with the MICROSOFT_BRAIN_DAMAGE option set, in order to disable support for the IDLE command. However, Entourage won't get new mail unless IDLE command support exists.

Note: the MICROSOFT_BRAIN_DAMAGE option no longer exists in modern versions, as the Outlook Express problem which it attempted to solve has been worked around in another way.

7.39 Why doesn't Entourage work at all?

It's hard to know. Entourage breaks almost every rule in the book for IMAP. It is highly instructive to do a packet trace on Entourage, as an example of how not to use IMAP. It does things like STATUS (MESSAGES) on the currently selected mailbox and re-fetching the same static data over and over again.

It seems that every time we understand what it is doing wrong in Entourage and come up with a workaround, we learn about something else that's broken.

Try building imapd with the ENTOURAGE_BRAIN_DAMAGE option set, in order to disable the diagnostic that occurs when doing STATUS on the currently selected mailbox.

7.40 Why doesn't Netscape Notify (NSNOTIFY.EXE) work at all?

This is a bug in NSNOTIFY; it doesn't handle unsolicited data from the server correctly.

Fortunately, there is no reason to use this program with IMAP; NSNOTIFY is a polling program to let you know when new mail has appeared in your maildrop. This is necessary with POP; but since IMAP dynamically announces new mail in the session you're better off (and will actually cause less load on the server!) keeping your mail reading program's IMAP session open and let IMAP do the notifying for you.

Consequently, the recommended fix for the NSNOTIFY problem is to delete the NSNOTIFY binary.

7.41 Why can't I connect via SSL to Eudora? It says the connection has been broken, and in the server syslogs I see "Command stream end of file".

There is a report that you can fix the problem by going into Eudora's advanced network configuration menu and increasing the network buffer size to 8192.

7.42 Sheesh. Aren't there any good IMAP clients out there?

Yes!

Pine is a wonderful client. It's fast, it uses IMAP well, and it generates text mail (life is too short to waste on HTML mail). Also, there are some really wonderful things in progress in the Pine world.

There are some good GUI clients out there, mostly from smaller vendors. Without naming names, look for the vendors who are active in the IMAP protocol development community, and their products.

Netscape, Eudora, and Outlook can be configured with enough effort to be good citizens and work well for users, but they can also be badly misconfigured, and often the misconfiguration is the default.

7.43 But wait! PC Pine (or other PC program build with c-client) crashes with the message incomplete SecBuffer exceeds maximum buffer size when I use SSL connections. This is a bug in c-client, right?

It's a bug in the Microsoft SChannel.DLL, which implements SSL. Microsoft admits it (albeit with an unstatement: "it's not fully RFC compliant"). The problem is that SChannel indicates that the maximum SSL packet data size is 5 bytes smaller than the actual maximum. Thus, any IMAP server which transmits a maximum sized SSL packet will not work with PC Pine or any other program which uses SChannel.

It can take a while for the problem to show up. The client has to do something that causes at least 16K of contiguous data. Many clients do partial fetching, which tends to reduce the number of cases where this can happen. However, all software which uses SChannel to support SSL is affected by this bug.

This problem does not affect UNIX code, since OpenSSL is used on UNIX.

This problem most recently showed up with the CommunigatePro IMAP server. They have an update which trims down their maximum contiguous data to less than 16K, in order to work around the problem.

This problem has also shown up with the Exchange IMAP server with UNIX clients (including Pine built with an older version of c-client) which sends full-sized 16K SSL packets. Modern c-client works around the problem by trimming down its maximum outgoing SSL packet size to 8K.

Microsoft has developed a hotfix for this bug. Look up MSKB article number 300562. Contrary to the article text which implies that this is a Pine issue, this bug also affect Microsoft Exchange server with *any* UNIX based client that transmits full-sized SSL payloads.

7.44 My qpopper users keep on getting the DON'T DELETE THIS MESSAGE -- FOLDER INTERNAL DATA if they also use Pine or IMAP. How can I fix this?

This is an incompatibility between qpopper and the c-client library used by Pine, imapd, and ipop[23]d.

Assuming that you want to continue using qpopper, look into qpopper's --enable-uw-kludge-flag configuration flag, which is documented as "check for and hide UW 'Folder Internal Data' messages".

The other alternative is to switch from qpopper to ipop3d.

7.45 Help! I installed the servers but I can't connect to them from my client!

Review the installation instructions carefully. Make sure that you have not skipped any of the steps. Make sure that you have made the correct entries in the configuration files; pay careful attention to the exact spelling of the service names and the path names. Make sure as well that you have properly restarted inetd.

If you have a system with Yellow Pages/NIS such as Solaris, have you updated the service names there as well as in /etc/services?

If you have a system with TCP wrappers, have you properly updated the TCP wrapper files (e.g. /etc/hosts.allow and /etc/hosts.deny) for the servers?

If you have a system which uses xinetd instead of inetd, have you made sure that you have made the correct corresponding xinetd changes for those services?

Try telneting to the server port (143 for IMAP, 110 for POP3). If you get a "refused" error, that probably means that you don't have the service set up in inetd.conf. If the connection opens and then closes with no message, the service is set up, but either the path name of the server binary in inetd.conf is wrong or your TCP wrappers are configured to deny access.

If you don't know how to make the corresponding changes to these files, seek the help of a local expert for your system.

7.46 Why do I get the message Can not authenticate to SMTP server: 421 SMTP connection went away! and why did this happen? There was also something about SECURITY PROBLEM: insecure server advertised AUTH=PLAIN

Some versions of qmail, including that running on mail.smtp.yahoo.com, disconnect the SMTP session if you fail to authenticate prior to attempting to transmit mail. An attempt to authenticate was made, but it failed because the server had already disconnected.

To work around this, you need to specify /user=... in the host name specification.

The SECURITY PROBLEM came about because the server advertised the AUTH=PLAIN SASL authentication mechanism outside of a TLS-encrypted session, in violation of RFC 4616. This message is just a warning, and in fact occurred after the server had disconnected.

7.47 Why do I get the message SMTP Authentication cancelled and why did this happen? There was also something about SECURITY PROBLEM: insecure server advertised AUTH=PLAIN

This is a bug in the SMTP server.

Some versions of qmail, including that running on mail.smtp.yahoo.com, have a bug in their implementation of SASL in their SMTP server, which renders it non-compliant with the standard.

If the client does not provide an initial response in the command line for an authentication mechanism whose profile does not have an initial challenge, qmail issues a bogus response:

 334 ok, go on

The problem is the "ok, go on". This violates RFC 4954's requirement that the text part in a 334 response be a BASE64 encoded string; in other words, it is a protocol syntax error.

In the case of AUTH=PLAIN, RFC 4422 (page 7) requires that the encoded string have no data. In other words, the appropropiate standards-compliant server response is "334" followed by a SPACE and a CRLF.

7.48 Why do I get the message Invalid base64 string when I try to authenticate to a Cyrus server?

This slightly misleading message is the way that a Cyrus server indicates that an authentication exchange was cancelled. It is not indicative of a bug or protocol violation.

The most common reason that this happens is if the Cyrus server offers Kerberos authentication, c-client is built with Kerberos support, but your client system is not within the Kerberos realm. In this case, the client code will try to authenticate via Kerberos, fail to get the Kerberos credentials, cancel the authentication attempt, and try the next available authentication technology.

8. Where to Go For Additional Information

8.1 Where can I go to ask questions?
8.2 I have some ideas for enhancements to IMAP. Where should I go?

If you have questions about the IMAP protocol, or want to participate in discussions of future directions of the IMAP protocol, the appropriate mailing list is imap-protocol@u.washington.edu. You can subscribe to this list via imap-protocol-request@u.washington.edu

If you have questions about this software, you can send me email directly or use the imap-uw@u.washington.edu mailing list. You can subscribe to this list via imap-uw-request@u.washington.edu

If you have general questions about the use of IMAP software (not specific to the UW IMAP toolkit) use the imap-use@u.washington.edu mailing list. You can subscribe to this list via imap-use-request@u.washington.edu

You must be a subscriber to post to these lists. As an alternative, you can use the comp.mail.imap newsgroup.

8.3 Where can I read more about IMAP and other email protocols?

We recommend Internet Email Protocols: A Developer's Guide, by Kevin Johnson, published by Addison Wesley, ISBN 0-201-43288-9.

8.4 Where can I find out more about setting up and administering an IMAP server?

We recommend Managing IMAP, by Dianna Mullet & Kevin Mullet, published by O'Reilly, ISBN 0-596-00012-X.

This book also has an excellent comparison of the UW and Cyrus IMAP servers.

Last Updated: 15 November 2007

Table of Contents