Small patch from the Sieve maintainer.
[exim.git] / doc / doc-txt / README.SIEVE
1 $Cambridge: exim/doc/doc-txt/README.SIEVE,v 1.10 2006/04/25 10:44:57 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-05.txt (next version of RFC 3028 that fixes
25 specification mistakes), the "envelope" test (3028bis), the "fileinto"
26 action (3028bis), the "copy" action (RFC 3894), the "vacation" action
27 (draft-ietf-sieve-vacation-05.txt) and the "i;ascii-numeric" comparator
28 extension (RFC 2244).
29
30 The Sieve filter is integrated in Exim and works very similar to the
31 Exim filter: Sieve scripts are recognized by the first line containing
32 "# sieve filter". When using "keep" or "fileinto" to save a mail into a
33 folder, the resulting string is available as the variable $address_file
34 in the transport that stores it. The following routers and transport
35 show a typical use of Sieve:
36
37 begin routers
38
39 localuser_verify:
40 driver = accept
41 domains = +localdomains
42 local_part_suffix = "-*"
43 local_part_suffix_optional
44 check_local_user
45 require_files = $home/.forward
46 verify_only = true
47
48 localuser_deliver:
49 driver = redirect
50 domains = +localdomains
51 local_part_suffix = "-*"
52 local_part_suffix_optional
53 sieve_subaddress = "${sg{$local_part_suffix}{^-}{}}"
54 sieve_useraddress = "$local_part"
55 check_local_user
56 require_files = $home/.forward
57 file = $home/.forward
58 check_ancestor
59 allow_filter
60 file_transport = localuser
61 reply_transport = vacation
62 sieve_vacation_directory = $home/mail/vacation
63 verify = false
64
65 begin transports
66
67 localuser:
68 driver = appendfile
69 file = ${if eq{$address_file}{inbox} \
70 {/var/mail/$local_part} \
71 {${if eq{${substr_0_1:$address_file}}{/} \
72 {$address_file} \
73 {$home/mail/$address_file} \
74 }} \
75 }
76 delivery_date_add
77 envelope_to_add
78 return_path_add
79 mode = 0600
80
81 vacation:
82 driver = autoreply
83
84 Absolute files are stored where specified, relative files are stored
85 relative to $home/mail and "inbox" goes to the standard mailbox location.
86 To enable "vacation", sieve_vacation_directory is set to the directory
87 where vacation databases are held (don't put anything else in that
88 directory) and point reply_transport to an autoreply transport.
89 Setting the Sieve useraddress and subaddress allows to use the subaddress
90 extension.
91
92
93 RFC Compliance
94
95 Exim requires the first line to be "# sieve filter". Of course the RFC
96 does not enforce that line. Don't expect examples to work without adding
97 it, though.
98
99 RFC 3028 requires using CRLF to terminate the end of a line.
100 The rationale was that CRLF is universally used in network protocols
101 to mark the end of the line. This implementation does not embed Sieve
102 in a network protocol, but uses Sieve scripts as part of the Exim MTA.
103 Since all parts of Exim use \n as newline character, this implementation
104 does, too. You can change this by defining the macro RFC_EOL at compile
105 time to enforce CRLF being used.
106
107 Sieve scripts can not contain NUL characters in strings, but mail
108 headers could contain MIME encoded NUL characters, which could never
109 be matched by Sieve scripts using exact comparisons. For that reason,
110 this implementation extends the Sieve quoted string syntax with \0
111 to describe a NUL character, violating \0 being the same as 0 in
112 RFC 3028.
113
114 The folder specified by "fileinto" must not contain the character
115 sequence ".." to avoid security problems. RFC 3028 does not specify the
116 syntax of folders apart from keep being equivalent to fileinto "INBOX".
117 This implementation uses "inbox" instead.
118
119 Sieve script errors currently cause that messages are silently filed into
120 "inbox". RFC 3028 requires that the user is notified of that condition.
121 This may be implemented in future by adding a header line to mails that
122 are filed into "inbox" due to an error in the filter.
123
124 The automatic replies generated by "vacation" do not contain an updated
125 "references" header field.
126
127
128 Semantics Of Keep
129
130 The keep command is equivalent to fileinto "inbox": It saves the
131 message and resets the implicit keep flag. It does not set the
132 implicit keep flag; there is no command to set it once it has
133 been reset.
134
135
136 Semantics Of Fileinto
137
138 RFC 3028 does not specify if "fileinto" tries to create a mail folder,
139 in case it does not exist. This implementation allows to configure
140 that aspect using the appendfile transport options "create_directory",
141 "create_file" and "file_must_exist". See the appendfile transport in
142 the Exim specification for details.
143
144
145 Allof And Anyof Test
146
147 RFC 3028 does not specify if these tests use shortcut/lazy evaluation.
148 Exim uses shortcut evaluation.
149
150
151 Action Reordering
152
153 RFC 3028 does not specify if actions may be executed out of order.
154 Exim may execute them out of order, e.g. messages may be filed to
155 folders or forwarded in a different order than specified, because
156 those actions only setup delivery, but do not execute it themselves.
157
158
159 Wildcard Matching
160
161 RFC 3028 is not exactly clear if comparators act on unicode characters
162 or on octets containing their UTF-8 representation. As it turns out,
163 many implementations go the second way. This does not make a difference
164 but for wildcard matching and octet-wise comparison. Working on unicode
165 means a dot matches a character. Working on UTF-8 means the dot matches
166 a single octet of a multi-octet sequence. For octet-wise comparisons,
167 working on UTF-8 means arbitrary byte sequences in headers can not be
168 matches, as they are rarely correct UTF-8 sequences and can thus not be
169 expressed as string literal. This implementation works on unicode, but
170 this may be changed in case RFC3028bis specifies this issue safe and sound.
171
172
173 Sieve Syntax And Semantics
174
175 RFC 3028 confuses syntax and semantics sometimes. It uses a generic
176 grammar as syntax for commands and tests and performs many checks during
177 semantic analysis. Syntax is specified by grammar rules, semantics
178 by natural language, despite the latter often talking about syntax.
179 The intention was to provide a framework for the syntax that describes
180 current commands as well as future extensions, and describing commands
181 by semantics.
182
183 The following replacement for section 8.2 gives two grammars, one for
184 the framework, and one for specific commands, thus removing most of the
185 semantic analysis. Since the parser can not parse unsupported extensions,
186 the result is strict error checking of any executed and not executed code
187 until "stop" is executed or the end of the script is reached.
188
189 8.2. Grammar
190
191 The atoms of the grammar are lexical tokens. White space or comments may
192 appear anywhere between lexical tokens, they are not part of the grammar.
193 The grammar is specified in ABNF with two extensions to describe tagged
194 arguments that can be reordered and grammar extensions: { } denotes a
195 sequence of symbols that may appear in any order. Example:
196
197 options = a b c
198 start = { options }
199
200 is equivalent to:
201
202 start = ( a b c ) / ( a c b ) / ( b a c ) / ( b c a ) / ( c a b ) / ( c b a )
203
204 The symbol =) is used to append to a rule:
205
206 start = a
207 start =) b
208
209 is equivalent to
210
211 start = a b
212
213 All Sieve commands, including extensions, MUST be words of the following
214 generic grammar with the start symbol "start". They SHOULD be specified
215 using a specific grammar, though.
216
217 argument = string-list / number / tag
218 arguments = *argument [test / test-list]
219 block = "{" commands "}"
220 commands = *command
221 string = quoted-string / multi-line
222 string-list = "[" string *("," string) "]" / string
223 test = identifier arguments
224 test-list = "(" test *("," test) ")"
225 command = identifier arguments ( ";" / block )
226 start = command
227
228 The basic Sieve commands are specified using the following grammar, which
229 language is a subset of the generic grammar above. The start symbol is
230 "start".
231
232 address-part = ":localpart" / ":domain" / ":all"
233 comparator = ":comparator" string
234 match-type = ":is" / ":contains" / ":matches"
235 string = quoted-string / multi-line
236 string-list = "[" string *("," string) "]" / string
237 address-test = "address" { [address-part] [comparator] [match-type] }
238 string-list string-list
239 test-list = "(" test *("," test) ")"
240 allof-test = "allof" test-list
241 anyof-test = "anyof" test-list
242 exists-test = "exists" string-list
243 false-test = "false"
244 true=test = "true"
245 header-test = "header" { [comparator] [match-type] }
246 string-list string-list
247 not-test = "not" test
248 relop = ":over" / ":under"
249 size-test = "size" relop number
250 block = "{" commands "}"
251 if-command = "if" test block *( "elsif" test block ) [ "else" block ]
252 stop-command = "stop" { stop-options } ";"
253 stop-options =
254 keep-command = "keep" { keep-options } ";"
255 keep-options =
256 discard-command = "discard" { discard-options } ";"
257 discard-options =
258 redirect-command = "redirect" { redirect-options } string ";"
259 redirect-options =
260 require-command = "require" { require-options } string-list ";"
261 require-options =
262 test = address-test / allof-test / anyof-test / exists-test
263 / false-test / true-test / header-test / not-test
264 / size-test
265 command = if-command / stop-command / keep-command
266 / discard-command / redirect-command
267 commands = *command
268 start = *require-command commands
269
270 The extensions "envelope" and "fileinto" are specified using the following
271 grammar extension.
272
273 envelope-test = "envelope" { [comparator] [address-part] [match-type] }
274 string-list string-list
275 test =/ envelope-test
276
277 fileinto-command = "fileinto" { fileinto-options } string ";"
278 fileinto-options =
279 command =/ fileinto-command
280
281 The extension "copy" is specified as:
282
283 fileinto-options =) ":copy"
284 redirect-options =) ":copy"
285
286
287 The i;ascii-numeric Comparator
288
289 RFC 2244 describes this comparator and specifies that non-numeric strings
290 are considered equal with an ordinal value higher than any numeric string.
291 Although not stated explicitly, this includes the empty string. A range
292 of at least 2^31 is required. This implementation does not limit the
293 range, because it does not convert numbers to binary representation
294 before comparing them.
295
296
297 The vacation extension
298
299 The extension "vacation" is specified using the following grammar
300 extension.
301
302 vacation-command = "vacation" { vacation-options } <reason: string>
303 vacation-options = [":days" number]
304 [":subject" string]
305 [":from" string]
306 [":addresses" string-list]
307 [":mime"]
308 [":handle" string]
309 command =/ vacation-command
310
311
312 Semantics Of ":mime"
313
314 The draft does not specify how strings using MIME entities are used
315 to compose messages. As a result, different implementations generate
316 different mails. The Exim Sieve implementation splits the reason into
317 header and body. It adds the header to the mail header and uses the body
318 as mail body. Be aware, that other imlementations compose a multipart
319 structure with the reason as only part. Both conform to the specification
320 (or lack thereof).
321
322
323 Semantics Of Not Using ":mime"
324
325 Sieve scripts are written in UTF-8, so is the reason string in this
326 case. This implementation adds MIME headers to indicate that. This
327 is not required by the vacation draft, which does not specify how
328 the UTF-8 reason is processed to compose the resulting message.
329
330
331 Default Subject
332
333 The draft specifies that the default message subject is "Auto: " plus
334 the old subject. Using this subject is dangerous, because many mailing
335 lists verify addresses by sending a secret key in the subject of a
336 message, asking to reply to the message for confirmation. Using the
337 default vacation subject confirms any subscription request of this kind,
338 allowing to subscribe a third party to any mailing list, either to annoy
339 the user or to declare spam as legitimate mail by proving to use opt-in.
340
341
342 Rate Limiting Responses
343
344 In absence of a handle, this implementation hashes the reason,
345 ":subject" option, ":mime" option and ":from" option and uses the hex
346 string representation as filename within the "sieve_vacation_directory"
347 to store the recipient addresses for this vacation parameter set.
348
349 The draft specifies that sites may define a minimum ":days" value than 1.
350 This implementation uses 1. The maximum value MUST greater than 7,
351 and SHOULD be greater than 30. This implementation uses a maximum of 31.
352
353 Vacation recipient address databases older than 31 days are automatically
354 removed. Users do not have to remove them manually when modifying their
355 scripts. Don't put anything but vacation databases in that directory
356 or you risk that it will be removed, too!
357
358
359 Global Reply Address Blacklist
360
361 The draft requires that each implementation offers a global black list
362 of addresses that will never be replied to. Exim offers this as option
363 "never_mail" in the autoreply transport.