Patch from the Sieve maintainer.
[exim.git] / doc / doc-txt / README.SIEVE
1 $Cambridge: exim/doc/doc-txt/README.SIEVE,v 1.11 2007/03/21 15:15:12 ph10 Exp $
2
3               Notes on the Sieve implementation for Exim
4
5 Exim Filter Versus Sieve Filter
6
7 Exim supports two incompatible filters: The traditional Exim filter and
8 the Sieve filter. Since Sieve is a extensible language, it is important
9 to understand "Sieve" in this context as "the specific implementation
10 of Sieve for Exim".
11
12 The Exim filter contains more features, such as variable expansion, and
13 better integration with the host environment, like external processes
14 and pipes.
15
16 Sieve is a standard for interoperable filters, defined in RFC 3028,
17 with multiple implementations around. If interoperability is important,
18 then there is no way around it.
19
20
21 Exim Implementation
22
23 The Exim Sieve implementation offers the core as defined by
24 draft-ietf-sieve-3028bis-10.txt (next version of RFC 3028 that
25 fixes specification mistakes), the "envelope" test (3028bis), the
26 "fileinto" action (3028bis), the "copy" parameter (RFC 3894), the
27 "vacation" action (draft-ietf-sieve-vacation-06), the "notify" action
28 (draft-ietf-sieve-notify-06.), the "i;ascii-numeric" comparator (RFC 2244)
29 and the subaddress parameter (draft-ietf-sieve-rfc3598bis-05).
30
31 The Sieve filter is integrated in Exim and works very similar to the
32 Exim filter: Sieve scripts are recognized by the first line containing
33 "# sieve filter".  When using "keep" or "fileinto" to save a mail into a
34 folder, the resulting string is available as the variable $address_file
35 in the transport that stores it.  The following routers and transport
36 show a typical use of Sieve:
37
38 begin routers
39
40 localuser_verify:
41   driver = accept
42   domains = +localdomains
43   local_part_suffix = "-*"
44   local_part_suffix_optional
45   check_local_user
46   require_files = $home/.forward
47   verify_only = true
48
49 localuser_deliver:
50   driver = redirect
51   domains = +localdomains
52   local_part_suffix = "-*"
53   local_part_suffix_optional
54   sieve_subaddress = "${sg{$local_part_suffix}{^-}{}}"
55   sieve_useraddress = "$local_part"
56   check_local_user
57   require_files = $home/.forward
58   file = $home/.forward
59   check_ancestor
60   allow_filter
61   file_transport = localuser
62   reply_transport = vacation
63   sieve_vacation_directory = $home/mail/vacation
64   verify = false
65
66 begin transports
67
68 localuser:
69   driver = appendfile
70   file = ${if eq{$address_file}{inbox} \
71               {/var/mail/$local_part} \
72               {${if eq{${substr_0_1:$address_file}}{/} \
73                     {$address_file} \
74                     {$home/mail/$address_file} \
75               }} \
76          }
77   delivery_date_add
78   envelope_to_add
79   return_path_add
80   mode = 0600
81
82 vacation:
83   driver = autoreply
84
85 Absolute files are stored where specified, relative files are stored
86 relative to $home/mail and "inbox" goes to the standard mailbox location.
87 To enable "vacation", sieve_vacation_directory is set to the directory
88 where vacation databases are held (don't put anything else in that
89 directory) and point reply_transport to an autoreply transport.
90 Setting the Sieve useraddress and subaddress allows to use the subaddress
91 extension.
92
93
94 RFC Compliance
95
96 Exim requires the first line to be "# sieve filter".  Of course the RFC
97 does not enforce that line.  Don't expect examples to work without adding
98 it, though.
99
100 RFC 3028 requires using CRLF to terminate the end of a line.
101 The rationale was that CRLF is universally used in network protocols
102 to mark the end of the line.  This implementation does not embed Sieve
103 in a network protocol, but uses Sieve scripts as part of the Exim MTA.
104 Since all parts of Exim use \n as newline character, this implementation
105 does, too.  You can change this by defining the macro RFC_EOL at compile
106 time to enforce CRLF being used.
107
108 Sieve scripts can not contain NUL characters in strings, but mail
109 headers could contain MIME encoded NUL characters, which could never
110 be matched by Sieve scripts using exact comparisons.  For that reason,
111 this implementation extends the Sieve quoted string syntax with \0
112 to describe a NUL character, violating \0 being the same as 0 in
113 RFC 3028.
114
115 The folder specified by "fileinto" must not contain the character
116 sequence ".." to avoid security problems.  RFC 3028 does not specify the
117 syntax of folders apart from keep being equivalent to fileinto "INBOX".
118 This implementation uses "inbox" instead.
119
120 Sieve script errors currently cause that messages are silently filed into
121 "inbox".  RFC 3028 requires that the user is notified of that condition.
122 This may be implemented in future by adding a header line to mails that
123 are filed into "inbox" due to an error in the filter.
124
125 The automatic replies generated by "vacation" do not contain an updated
126 "references" header field.
127
128
129 Semantics Of Keep
130
131 The keep command is equivalent to fileinto "inbox": It saves the
132 message and resets the implicit keep flag.  It does not set the
133 implicit keep flag; there is no command to set it once it has
134 been reset.
135
136
137 Semantics Of Fileinto
138
139 RFC 3028 does not specify if "fileinto" tries to create a mail folder,
140 in case it does not exist.  This implementation allows to configure
141 that aspect using the appendfile transport options "create_directory",
142 "create_file" and "file_must_exist".  See the appendfile transport in
143 the Exim specification for details.
144
145
146 Allof And Anyof Test
147
148 RFC 3028 does not specify if these tests use shortcut/lazy evaluation.
149 Exim uses shortcut evaluation.
150
151
152 Action Reordering
153
154 RFC 3028 does not specify if actions may be executed out of order.
155 Exim may execute them out of order, e.g. messages may be filed to
156 folders or forwarded in a different order than specified, because
157 those actions only setup delivery, but do not execute it themselves.
158
159
160 Wildcard Matching
161
162 RFC 3028 is not exactly clear if comparators act on unicode characters
163 or on octets containing their UTF-8 representation.  As it turns out,
164 many implementations go the second way.  This does not make a difference
165 but for wildcard matching and octet-wise comparison.  Working on unicode
166 means a dot matches a character.  Working on UTF-8 means the dot matches
167 a single octet of a multi-octet sequence.  For octet-wise comparisons,
168 working on UTF-8 means arbitrary byte sequences in headers can not be
169 matches, as they are rarely correct UTF-8 sequences and can thus not be
170 expressed as string literal.  This implementation works on unicode, but
171 this may be changed in case RFC3028bis specifies this issue safe and sound.
172
173
174 Sieve Syntax And Semantics
175
176 RFC 3028 confuses syntax and semantics sometimes.  It uses a generic
177 grammar as syntax for commands and tests and performs many checks during
178 semantic analysis.  Syntax is specified by grammar rules, semantics
179 by natural language, despite the latter often talking about syntax.
180 The intention was to provide a framework for the syntax that describes
181 current commands as well as future extensions, and describing commands
182 by semantics.
183
184 The following replacement for section 8.2 gives two grammars, one for
185 the framework, and one for specific commands, thus removing most of the
186 semantic analysis.  Since the parser can not parse unsupported extensions,
187 the result is strict error checking of any executed and not executed code
188 until "stop" is executed or the end of the script is reached.
189
190 8.2. Grammar
191
192 The atoms of the grammar are lexical tokens.  White space or comments may
193 appear anywhere between lexical tokens, they are not part of the grammar.
194 The grammar is specified in ABNF with two extensions to describe tagged
195 arguments that can be reordered and grammar extensions: { } denotes a
196 sequence of symbols that may appear in any order.  Example:
197
198   options = a b c
199   start   = { options }
200
201 is equivalent to:
202
203   start   =  ( a b c ) / ( a c b ) / ( b a c ) / ( b c a ) / ( c a b ) / ( c b a )
204
205 The symbol =) is used to append to a rule:
206
207   start =  a
208   start =) b
209
210 is equivalent to
211
212   start =  a b
213
214 All Sieve commands, including extensions, MUST be words of the following
215 generic grammar with the start symbol "start".  They SHOULD be specified
216 using a specific grammar, though.
217
218    argument        = string-list / number / tag
219    arguments       = *argument [test / test-list]
220    block           = "{" commands "}"
221    commands        = *command
222    string          = quoted-string / multi-line
223    string-list     = "[" string *("," string) "]" / string
224    test            = identifier arguments
225    test-list       = "(" test *("," test) ")"
226    command         = identifier arguments ( ";" / block )
227    start           = command
228
229 The basic Sieve commands are specified using the following grammar, which
230 language is a subset of the generic grammar above.  The start symbol is
231 "start".
232
233   address-part     =  ":localpart" / ":domain" / ":all"
234   comparator       =  ":comparator" string
235   match-type       =  ":is" / ":contains" / ":matches"
236   string           =  quoted-string / multi-line
237   string-list      =  "[" string *("," string) "]" / string
238   address-test     =  "address" { [address-part] [comparator] [match-type] }
239                       string-list string-list
240   test-list        =  "(" test *("," test) ")"
241   allof-test       =  "allof" test-list
242   anyof-test       =  "anyof" test-list
243   exists-test      =  "exists" string-list
244   false-test       =  "false"
245   true=test        =  "true"
246   header-test      =  "header" { [comparator] [match-type] }
247                       string-list string-list
248   not-test         =  "not" test
249   relop            =  ":over" / ":under"
250   size-test        =  "size" relop number
251   block            =  "{" commands "}"
252   if-command       =  "if" test block *( "elsif" test block ) [ "else" block ]
253   stop-command     =  "stop" { stop-options } ";"
254   stop-options     =
255   keep-command     =  "keep" { keep-options } ";"
256   keep-options     =
257   discard-command  =  "discard" { discard-options } ";"
258   discard-options  =
259   redirect-command =  "redirect" { redirect-options } string ";"
260   redirect-options =
261   require-command  =  "require" { require-options } string-list ";"
262   require-options  =
263   test             =  address-test / allof-test / anyof-test / exists-test
264                       / false-test / true-test / header-test / not-test
265                       / size-test
266   command          =  if-command / stop-command / keep-command
267                       / discard-command / redirect-command
268   commands         =  *command
269   start            =  *require-command commands
270
271 The extensions "envelope" and "fileinto" are specified using the following
272 grammar extension.
273
274   envelope-test    =  "envelope" { [comparator] [address-part] [match-type] }
275                       string-list string-list
276   test             =/ envelope-test
277
278   fileinto-command =  "fileinto" { fileinto-options } string ";"
279   fileinto-options =
280   command          =/ fileinto-command
281
282 The extension "copy" is specified as:
283
284   fileinto-options =) ":copy"
285   redirect-options =) ":copy"
286
287
288 The i;ascii-numeric Comparator
289
290 RFC 2244 describes this comparator and specifies that non-numeric strings
291 are considered equal with an ordinal value higher than any numeric string.
292 Although not stated explicitly, this includes the empty string.  A range
293 of at least 2^31 is required.  This implementation does not limit the
294 range, because it does not convert numbers to binary representation
295 before comparing them.
296
297
298 The vacation extension
299
300 The extension "vacation" is specified using the following grammar
301 extension.
302
303   vacation-command =  "vacation" { vacation-options } <reason: string>
304   vacation-options =  [":days" number]
305                       [":subject" string]
306                       [":from" string]
307                       [":addresses" string-list]
308                       [":mime"]
309                       [":handle" string]
310   command          =/ vacation-command
311
312
313 Semantics Of ":mime"
314
315 The draft does not specify how strings using MIME entities are used
316 to compose messages.  As a result, different implementations generate
317 different mails.  The Exim Sieve implementation splits the reason into
318 header and body.  It adds the header to the mail header and uses the body
319 as mail body.  Be aware, that other imlementations compose a multipart
320 structure with the reason as only part.  Both conform to the specification
321 (or lack thereof).
322
323
324 Semantics Of Not Using ":mime"
325
326 Sieve scripts are written in UTF-8, so is the reason string in this
327 case.  This implementation adds MIME headers to indicate that.  This
328 is not required by the vacation draft, which does not specify how
329 the UTF-8 reason is processed to compose the resulting message.
330
331
332 Default Subject
333
334 The draft specifies that the default message subject is "Auto: " plus
335 the old subject.  Using this subject is dangerous, because many mailing
336 lists verify addresses by sending a secret key in the subject of a
337 message, asking to reply to the message for confirmation.  Using the
338 default vacation subject confirms any subscription request of this kind,
339 allowing to subscribe a third party to any mailing list, either to annoy
340 the user or to declare spam as legitimate mail by proving to use opt-in.
341
342
343 Rate Limiting Responses
344
345 In absence of a handle, this implementation hashes the reason,
346 ":subject" option, ":mime" option and ":from" option and uses the hex
347 string representation as filename within the "sieve_vacation_directory"
348 to store the recipient addresses for this vacation parameter set.
349
350 The draft specifies that sites may define a minimum ":days" value than 1.
351 This implementation uses 1.  The maximum value MUST greater than 7,
352 and SHOULD be greater than 30.  This implementation uses a maximum of 31.
353
354 Vacation recipient address databases older than 31 days are automatically
355 removed.  Users do not have to remove them manually when modifying their
356 scripts.  Don't put anything but vacation databases in that directory
357 or you risk that it will be removed, too!
358
359
360 Global Reply Address Blacklist
361
362 The draft requires that each implementation offers a global black list
363 of addresses that will never be replied to.  Exim offers this as option
364 "never_mail" in the autoreply transport.
365
366
367 The enotify extension
368
369 The extension "enotify" is specified using the following grammar
370 extension.
371
372   notify-command =  "notify" { notify-options } <method: string>
373   notify-options =  [":from" string]
374                     [":importance" <"1" / "2" / "3">]
375                     [":options" 1*(string-list / number)]
376                     [":message" string]
377
378   command          =/ notify-command
379
380   valid_notify_method = "valid_notify_method"
381                         <notification-uris: string-list>
382
383   test             =/ valid_notify_method
384
385 Only the mailto URI scheme is implemented.