Cyrus Carnegie Mellon

Project Cyrus Archives: Server Overview and Concepts
Computing Services
Home
ASG Home
What's New
Cyrus Wiki
Employment
Contact Us
Google

This document has not been reviewed in detail since 1997. However, the technical content is still basically valid (as of 2002).

This section describes mailbox namespace. The Cyrus IMAP server presents mailboxes using the netnews namespace convention. While mailbox names are case-insensitive, the case used when creating a mailbox is preserved. A mailbox name may not start or end with a "." character, nor may it contain two "." characters in a row.

In the initial implementation, non-ASCII characters, shell metacharacters, and "/" are not permitted in mailbox names. Later versions of the server may make the set of permissible characters configurable at each site.

All personal mailboxes for user "bovik" begin with the string "user.bovik.". The mailbox "user.bovik" is a special case. From user bovik's point of view, the mailbox user.bovik is named INBOX. If the access control list of the user.bovik mailbox permits other users to see the mailbox, it appears to the other users as user.bovik.

The mailbox user.bovik is referred to in this document as user bovik's INBOX.

Administrators create and delete users by creating and deleting the users' INBOX. If a user has an INBOX, then they are allowed to subscribe to mailboxes. Only users without dots in their userid are permitted to have an INBOX. (A user with a dot in their userid would be able to login but would not be able to receive mail.)

When an administrator deletes a user's INBOX, all of the user's personal mailboxes are deleted as well.

With the one notable exception of INBOX, all mailbox names are system-wide--they refer to the same mailbox regardless of the user. Access control lists determine which users can access or see which mailboxes.

In contexts which permit relative mailbox names, the mailbox namespace works as follows:

Thus, if you are working with folder names and the top of the hierarchy is named cmu., the name comp.infosystems.www resolves to comp.infosystems.www and the name .comp.infosystems.www resolves to cmu.comp.infosystems.www. Access

This section describes Access Control Lists and access rights.

Access to each mailbox is controlled by each mailbox's access control list. Access Control Lists (ACLs), provide a powerful mechanism for specifying the users or groups of users who have permission to access the mailboxes.

An ACL is a list of zero or more entries. Each entry has an identifier and a set of rights. The identifier specifies the user or group of users for which the entry applies. The set of rights is one or more letters or digits, each letter or digit conferring a particular privilege.

The defined rights are:

You can combine access rights in different ways. For example:

lrs
The user can read the mailbox.

lrsp
The user can read the mailbox, and can post to it through the delivery system. Most delivery systems do not provide authentication, so the p right usually has meaning only for the "anonymous" user.

lr
The user can see the mailbox and can read it, but the server does not preserve the "Seen" and "Recent" flags. This set of rights is primarily useful for "anonymous IMAP."

rs
The user can read the mailbox and the server preserves the "Seen" and "Recent" flags, but the mailbox is not visible to the user through the various mailbox listing commands. The user must know the name of the mailbox to be able to access it.

lrsip
The user can read and append to the mailbox, either through IMAP, or through the delivery system.

The identifier part of an ACL entry specifies the user or group for which the entry applies. The meaning of an identifier usually depends on the authorization mechanism being used.

With any authorization mechanism, two special identifiers are defined. The identifier anonymous refers to the anonymous, or unauthenticated user. The identifier anyone refers to all users, including the anonymous user.

The Cyrus IMAP server comes with two authorization mechanisms, one for use with Unix /etc/passwd files, one for use with Kerberos.

In the Unix authorization mechanism, identifiers are either a userid listed in /etc/passwd or the string group: followed by a group listed in /etc/group. Thus:

    root                Refers to the user root
    group:staff         Refers to the group staff
Using the Kerberos authorization mechanism, identifiers are of the form:

principal.instance@realm

If .instance is omitted, it defaults to the null string. If @realm is omitted, it defaults to the local realm.

Principal, instance, or realm may be replaced by the * wildcard.

None of the above refer to the anonymous user.

The file /etc/krb.equiv contains mappings between Kerberos principals. The file contains zero or more lines, each containing two fields. Any identity matching the first field of a line is changed to the second identity during canonicalization. For example, a line in /etc/krb.equiv of:

bovik@REMOTE.COM bovik will cause the identity bovik@REMOTE.COM to be treated as if it were the local principal bovik.

A site may wish to write their own authorization mechanism, perhaps to implement a local group mechanism.

Any of the above defined identifiers may be prefixed with a "-" character. The associated rights are then removed from that identifier. These are referred to as "negative rights."

To calculate the set of rights granted to a user, the server first calculates the union of all of the rights granted to the user and to all groups the user is a member of. The server then calculates and removes the union of all the negative rights granted to the user and to all groups the user is a member of.

For example, given the ACL:

   anyone       lrsp
   fred         lwi
   -anonymous   s

The user fred will be granted the rights lrswip and the anonymous user will be granted the rights lrp.

Regardless of the ACL on a mailbox, users who are listed in the "admins" configuration option in /etc/imapd.conf implicitly have the l and a rights on all mailboxes. Users also implicitly have the l and a rights on their INBOX and all of their personal mailboxes.

When a mailbox is created, its ACL starts off with a copy of the ACL of its closest parent mailbox. When a user is created, the ACL on the user's INBOX starts off with a single entry granting all rights to the user. When a non-user mailbox is created and does not have a parent, its ACL is initialized to the value of the defaultacl option in /etc/imapd.conf

This section discusses different types of logins that can be used with IMAP.

The Cyrus IMAP server comes with two login authentication mechanisms, one for use with Kerberos, one for use with Unix /etc/passwd files. Since authentication is closely tied with authorization, the login authentication mechanism chosen must be compatible with the authorization mechanism chosen.

Regardless of the login authentication mechanism chosen, the server can support anonymous login. If the allowanonymouslogin option in /etc/imapd.conf is turned on, then the server will permit plaintext password logins using the user anonymous and any password.

The Kerberos mechanism supports both the KERBEROS_V4 authentication mechanism and plaintext logins, verifying information using Kerberos version 4. The mechanism requires that a srvtab file exist in the location given in the srvtab configuration option. The srvtab file must be readable by the Cyrus server and must contain a imap.<host>@<realm> service key, where <host> is the first component of the server's host name and <realm> is the server's Kerberos realm. Plaintext passwords are verified by obtaining a ticket for the server's Kerberos identity, to protect against Kerberos server spoofing attacks.

The server will permit logins by identities in the local realm and identities in the realms listed in the loginrealms option in /etc/imapd.conf.

The file /etc/krb.equiv contains mappings between Kerberos principals. The file contains zero or more lines, each containing two fields. Any identity matching the first field of a line is permitted to log in as the identity in the second field.

The imap server identities for the hosts listed in the imspservers configuration option are permitted to log in as any user. This permits IMSP servers to perform commands for their clients by proxy.

If the loginuseacl configuration option is turned on, than any Kerberos identity that is granted the a right on the user's INBOX is permitted to log in as that user.

The Unix mechanism permits plaintext logins, verifying the supplied user and password against information in the Unix passwd file.

Quotas allow server administrators to limit resources used by hierarchies of mailboxes on the server.

The Cyrus IMAP server supports quotas on storage, which is defined as the number of bytes of the relevant RFC-822 messages, in kilobytes. Each copy of a message is counted independently, even when the server can conserve disk space use by making hard links to message files. The additional disk space overhead used by mailbox index and cache files is not charged against a quota.

Quotas are applied to quota roots, which can be at any level of the mailbox hierarchy. Quota roots need not also be mailboxes.

Quotas on a quota root apply to the sum of the usage of any mailbox at that level and any sub-mailboxes of that level that are not underneath a quota root on a sub-hierarchy. This means that each mailbox is limited by at most one quota root.

For example, if the mailboxes

   user.bovik
   user.bovik.list.imap
   user.bovik.list.info-cyrus
   user.bovik.saved
   user.bovik.todo

exist and the quota roots

   user.bovik 
   user.bovik.list
   user.bovik.saved

exist, then the quota root user.bovik applies to the mailboxes user.bovik and user.bovik.todo; the quota root user.bovik.list applies to the mailboxes user.bovik.list.imap and user.bovik.list.info-cyrus; and the quota root user.bovik.saved applies to the mailbox user.bovik.saved.

Quota roots are created automatically when they are mentioned in the setquota command. Quota roots may not be deleted through the protocol, see Removing Quota Roots for instructions on how to delete them.

Normally, in order for a message to be inserted into a mailbox, the quota root for the mailbox must have enough unused storage so that inserting the message will not cause the block quota to go over the limit.

Mail delivery is a special case. In order for a message to be delivered to a mailbox, the quota root for the mailbox must not have usage that is over the limit. If the usage is not over the limit, then one message may be delivered regardless of its size. This puts the mailbox's usage over the quota, causing a user to be informed of the problem and permitting them to correct it. If delivery were not permitted in this case, the user would have no practical way of knowing that there was mail that could not be delivered.

If the usage is over the limit, then the mail delivery will fail with a temporary error. This will cause the delivery system to re-attempt delivery for a couple of days (permitting the user time to notice and correct the problem) and then return the mail to the sender.

When a user selects a mailbox whose quota root has usage that is close to or over the limit and the user has dd rights on the mailbox, the server will issue an alert notifying the user that usage is close to or over the limit. The threshold of usage at which the server will issue quota warnings is set by the quotawarn configuration option.

The server only issues warnings when the user has d rights because only users with d rights are capable of correcting the problem.

Quota roots are independent of
partitions. A single quota root can apply to mailboxes in different partitions.

Partitions allow administrators to store different mailboxes in different parts of the Unix filesystem. This is intended to be used to allow hierarchies of mailboxes to be spread across multiple disks.

When an administrator creates a new mailbox, the name of the partition for the mailbox may be specified using an optional second argument to the create command. Non-administrators are not permitted to specify the partition of a mailbox. If the partition is not specified, then the mailbox inherits the partition of its most immediate parent mailbox. If the mailbox has no parent, it gets the partition specified in the defaultpartition configuration option.

The optional second argument to the create command can usually be given only when using a specialized administrative client such as cyradm.

An administrator may change the partition of a mailbox by using the rename command with an optional third argument. When a third argument to rename is given, the first and second arguments can be the same--this changes the partition of a mailbox without changing its name. If a third argument to rename is not given and the first argument is not INBOX, the partition of a mailbox does not change. If a third argument to rename is not given and the first argument is INBOX, the newly created mailbox gets the same partition it would get from the create command.

This section contains an overview of concepts related to exporting news groups through the IMAP server.

The partition named news is reserved for netnews and cannot be used for any other purpose.

The IMAP server can export netnews newsgroups as IMAP mailboxes. This is done by creating a partition named news which points to the news spool directory and arranging for the IMAP server's auxiliary databases to be maintained. The server software includes programs for integrating with the INN netnews server. The following information assumes working knowledge of the INN netnews server.

The collectnews program takes as input a list of article file path names relative to the news spool directory. It then adds those messages to the auxiliary databases. If collectnews encounters a newsgroup that does not have a corresponding IMAP mailbox listed in the mailboxes file (described later), it adds it there.

The collectnews program is normally run from cron as the cyrus user. It is given as input a news overview feed. Since collectnews needs to write the auxiliary databases in the news spool area, the cyrus user must be in the news group and the news group must be able to write to the news spool area.

The rmnews program takes a list of article pathnames relative to the news spool directory. It removes (expunges) any mention of those messages from the auxiliary databases and unlinks the article files. The rmnews program is used to remove both expired and canceled articles.

Rmnews takes the same input as the fastrm program. It is normally run to clean up after expire by invoking news.daily with the delayrm argument and having RMPROC in expirerm changed from fastrm to rmnews.

Rmnews is also normally run from cron as the cyrus user to remove canceled and superseded articles. It is given as input a special cancel feed. Patches to INN to add support for a cancel feed are included in the Cyrus IMAP server distribution.

As rmnews is invoked at expire time by the news user and must be able to write to the auxiliary databases, it must be installed with set-uid to the cyrus user and set-gid to the news group.

[[Removes the entry in mailboxes for removed newsgroups. Not written yet.]]

[[defer until implemented]]

The Cyrus IMAP server software comes with a compatible POP3 server. Due to limitations in the POP3 protocol, the server can only access a user's INBOX and only one instance of a POP3 server may exist for any one user at any time. While a POP3 server has a user's INBOX open, expunge operations from any concurrent IMAP session will fail.

When Kerberos login authentication is being used, the POP3 server uses the server identity pop.host@realm instead of imap.host@realm, where host is the first component of the server's host name and realm is the server's Kerberos realm. When the POP3 server is invoked with the -k switch, the server exports MIT's KPOP protocol instead of generic POP3.

The Cyrus IMAP server software sends log messages to the local6 syslog facility. The severity levels used are:

This section describes the various databases used by the Cyrus IMAP server software and what can be done to recover from various inconsistencies in these databases.

The largest database is the mailbox directory. Each mailbox directory contains the following files:

message files
There is one file per message, containing the message in RFC 822 format. Lines in the message are separated by CRLF, not just LF. The file name of each message is the message's UID followed by a dot (.).

In netnews newsgroups, the message files instead follow the format and naming conventions imposed by the netnews software.

cyrus.header
This file contains a magic number and variable-length information about the mailbox itself.

cyrus.index
This file contains fixed-length information about the mailbox itself and each message in the mailbox.

cyrus.cache
This file contans variable-length information about each message in the mailbox.

cyrus.seen
This file contains variable-length state information about each reader of the mailbox who has s permissions.

The reconstruct program can be used to recover from corruption in mailbox directories. If reconstruct can find existing header and index files, it attempts to preserve any data in them that is not derivable from the message files themselves. The state reconstruct attempts to preserve includes the flag names, flag state, and internal date. Reconstruct derives all other information from the message files.

An administrator may recover from a damaged disk by restoring message files from a backup and then running reconstruct to regenerate what it can of the other files.

The reconstruct program does not adjust the quota usage recorded in any quota root files. After running reconstruct, it is advisable to run quota -f (described below) in order to fix the quota root files.

The subdirectory quota of the configuration directory (specified in the configdirectory configuration option) contains one file per quota root, with the file name being the name of the quota root. These files store the quota usage and limits of each of the quota roots.

The quota program, when invoked with the -f switch, recalculates the quota root of each mailbox and the quota usage of each quota root.

To remove a quota root, remove the quota root's file. Then run quota -f to make the quota files consistent again.

The mailboxes file in the configuration directory is the most critical file in the entire Cyrus IMAP system. It contains a sorted list of each mailbox on the server, along with the mailboxes quota root and ACL. Because the ACL is security-critical information that cannot be reconstructed from information stored elsewhere, there is no program to recover from a damaged mailboxes file.

To protect the contents of the mailboxes, we suggest making frequent, even hourly backups of the mailboxes file to some other part of the disk.

The subdirectory user of the configuration directory contains user subscriptions. There is one file per user, with a filename of the userid followed by .sub. Each file contains a sorted list of subscribed mailboxes.

There is no program to recover from damaged subscription files. A site may recover from lost files simply by restoring from backups.

Many objects in the configuration directory are discussed in the Database Recovery section. This section documents two other directories that reside in the configuration directory.

The subdirectory log under the configuration directory permits administrators to keep protocol telemetry logs on a per-user basis.

If a subdirectory of log exists with the same name as a user, the IMAP and POP3 servers will keep a telemetry log of protocol sessions authenticating as that user. The telemetry log is stored in the subdirectory with a filename of the server process-id and starts with the first command following authentication.

The subdirectory proc under the configuration directory contains one file per active server process. The file name is the ASCII representation of the process id and the file contains the following tab-separated fields:

The file may contain arbitrary characters after the first newline character.

The proc subdirectory is normally be cleaned out on server reboot.

The following sites contain additional information about IMAP: