14199
|
1 |
(* Title: HOL/Auth/SET/MessageSET
|
|
2 |
ID: $Id$
|
|
3 |
Authors: Giampaolo Bella, Fabio Massacci, Lawrence C Paulson
|
|
4 |
*)
|
|
5 |
|
|
6 |
header{*The Message Theory, Modified for SET*}
|
|
7 |
|
|
8 |
theory MessageSET = NatPair:
|
|
9 |
|
|
10 |
subsection{*General Lemmas*}
|
|
11 |
|
|
12 |
text{*Needed occasionally with @{text spy_analz_tac}, e.g. in
|
|
13 |
@{text analz_insert_Key_newK}*}
|
|
14 |
|
|
15 |
lemma Un_absorb3 [simp] : "A \<union> (B \<union> A) = B \<union> A"
|
|
16 |
by blast
|
|
17 |
|
|
18 |
text{*Collapses redundant cases in the huge protocol proofs*}
|
|
19 |
lemmas disj_simps = disj_comms disj_left_absorb disj_assoc
|
|
20 |
|
|
21 |
text{*Effective with assumptions like @{term "K \<notin> range pubK"} and
|
|
22 |
@{term "K \<notin> invKey`range pubK"}*}
|
|
23 |
lemma notin_image_iff: "(y \<notin> f`I) = (\<forall>i\<in>I. f i \<noteq> y)"
|
|
24 |
by blast
|
|
25 |
|
|
26 |
text{*Effective with the assumption @{term "KK \<subseteq> - (range(invKey o pubK))"} *}
|
|
27 |
lemma disjoint_image_iff: "(A <= - (f`I)) = (\<forall>i\<in>I. f i \<notin> A)"
|
|
28 |
by blast
|
|
29 |
|
|
30 |
|
|
31 |
|
|
32 |
types
|
|
33 |
key = nat
|
|
34 |
|
|
35 |
consts
|
|
36 |
all_symmetric :: bool --{*true if all keys are symmetric*}
|
|
37 |
invKey :: "key=>key" --{*inverse of a symmetric key*}
|
|
38 |
|
|
39 |
specification (invKey)
|
|
40 |
invKey [simp]: "invKey (invKey K) = K"
|
|
41 |
invKey_symmetric: "all_symmetric --> invKey = id"
|
|
42 |
by (rule exI [of _ id], auto)
|
|
43 |
|
|
44 |
|
|
45 |
text{*The inverse of a symmetric key is itself; that of a public key
|
|
46 |
is the private key and vice versa*}
|
|
47 |
|
|
48 |
constdefs
|
|
49 |
symKeys :: "key set"
|
|
50 |
"symKeys == {K. invKey K = K}"
|
|
51 |
|
|
52 |
text{*Agents. We allow any number of certification authorities, cardholders
|
|
53 |
merchants, and payment gateways.*}
|
|
54 |
datatype
|
|
55 |
agent = CA nat | Cardholder nat | Merchant nat | PG nat | Spy
|
|
56 |
|
|
57 |
text{*Messages*}
|
|
58 |
datatype
|
|
59 |
msg = Agent agent --{*Agent names*}
|
|
60 |
| Number nat --{*Ordinary integers, timestamps, ...*}
|
|
61 |
| Nonce nat --{*Unguessable nonces*}
|
|
62 |
| Pan nat --{*Unguessable Primary Account Numbers (??)*}
|
|
63 |
| Key key --{*Crypto keys*}
|
|
64 |
| Hash msg --{*Hashing*}
|
|
65 |
| MPair msg msg --{*Compound messages*}
|
|
66 |
| Crypt key msg --{*Encryption, public- or shared-key*}
|
|
67 |
|
|
68 |
|
|
69 |
(*Concrete syntax: messages appear as {|A,B,NA|}, etc...*)
|
|
70 |
syntax
|
|
71 |
"@MTuple" :: "['a, args] => 'a * 'b" ("(2{|_,/ _|})")
|
|
72 |
|
|
73 |
syntax (xsymbols)
|
|
74 |
"@MTuple" :: "['a, args] => 'a * 'b" ("(2\<lbrace>_,/ _\<rbrace>)")
|
|
75 |
|
|
76 |
translations
|
|
77 |
"{|x, y, z|}" == "{|x, {|y, z|}|}"
|
|
78 |
"{|x, y|}" == "MPair x y"
|
|
79 |
|
|
80 |
|
|
81 |
constdefs
|
|
82 |
nat_of_agent :: "agent => nat"
|
|
83 |
"nat_of_agent == agent_case (curry nat2_to_nat 0)
|
|
84 |
(curry nat2_to_nat 1)
|
|
85 |
(curry nat2_to_nat 2)
|
|
86 |
(curry nat2_to_nat 3)
|
|
87 |
(nat2_to_nat (4,0))"
|
|
88 |
--{*maps each agent to a unique natural number, for specifications*}
|
|
89 |
|
|
90 |
text{*The function is indeed injective*}
|
|
91 |
lemma inj_nat_of_agent: "inj nat_of_agent"
|
|
92 |
by (simp add: nat_of_agent_def inj_on_def curry_def
|
|
93 |
nat2_to_nat_inj [THEN inj_eq] split: agent.split)
|
|
94 |
|
|
95 |
|
|
96 |
constdefs
|
|
97 |
(*Keys useful to decrypt elements of a message set*)
|
|
98 |
keysFor :: "msg set => key set"
|
|
99 |
"keysFor H == invKey ` {K. \<exists>X. Crypt K X \<in> H}"
|
|
100 |
|
|
101 |
subsubsection{*Inductive definition of all "parts" of a message.*}
|
|
102 |
|
|
103 |
consts parts :: "msg set => msg set"
|
|
104 |
inductive "parts H"
|
|
105 |
intros
|
|
106 |
Inj [intro]: "X \<in> H ==> X \<in> parts H"
|
|
107 |
Fst: "{|X,Y|} \<in> parts H ==> X \<in> parts H"
|
|
108 |
Snd: "{|X,Y|} \<in> parts H ==> Y \<in> parts H"
|
|
109 |
Body: "Crypt K X \<in> parts H ==> X \<in> parts H"
|
|
110 |
|
|
111 |
|
|
112 |
(*Monotonicity*)
|
|
113 |
lemma parts_mono: "G<=H ==> parts(G) <= parts(H)"
|
|
114 |
apply auto
|
|
115 |
apply (erule parts.induct)
|
|
116 |
apply (auto dest: Fst Snd Body)
|
|
117 |
done
|
|
118 |
|
|
119 |
|
|
120 |
subsubsection{*Inverse of keys*}
|
|
121 |
|
|
122 |
(*Equations hold because constructors are injective; cannot prove for all f*)
|
|
123 |
lemma Key_image_eq [simp]: "(Key x \<in> Key`A) = (x\<in>A)"
|
|
124 |
by auto
|
|
125 |
|
|
126 |
lemma Nonce_Key_image_eq [simp]: "(Nonce x \<notin> Key`A)"
|
|
127 |
by auto
|
|
128 |
|
|
129 |
lemma Cardholder_image_eq [simp]: "(Cardholder x \<in> Cardholder`A) = (x \<in> A)"
|
|
130 |
by auto
|
|
131 |
|
|
132 |
lemma CA_image_eq [simp]: "(CA x \<in> CA`A) = (x \<in> A)"
|
|
133 |
by auto
|
|
134 |
|
|
135 |
lemma Pan_image_eq [simp]: "(Pan x \<in> Pan`A) = (x \<in> A)"
|
|
136 |
by auto
|
|
137 |
|
|
138 |
lemma Pan_Key_image_eq [simp]: "(Pan x \<notin> Key`A)"
|
|
139 |
by auto
|
|
140 |
|
|
141 |
lemma Nonce_Pan_image_eq [simp]: "(Nonce x \<notin> Pan`A)"
|
|
142 |
by auto
|
|
143 |
|
|
144 |
lemma invKey_eq [simp]: "(invKey K = invKey K') = (K=K')"
|
|
145 |
apply safe
|
|
146 |
apply (drule_tac f = invKey in arg_cong, simp)
|
|
147 |
done
|
|
148 |
|
|
149 |
|
|
150 |
subsection{*keysFor operator*}
|
|
151 |
|
|
152 |
lemma keysFor_empty [simp]: "keysFor {} = {}"
|
|
153 |
by (unfold keysFor_def, blast)
|
|
154 |
|
|
155 |
lemma keysFor_Un [simp]: "keysFor (H \<union> H') = keysFor H \<union> keysFor H'"
|
|
156 |
by (unfold keysFor_def, blast)
|
|
157 |
|
|
158 |
lemma keysFor_UN [simp]: "keysFor (\<Union>i\<in>A. H i) = (\<Union>i\<in>A. keysFor (H i))"
|
|
159 |
by (unfold keysFor_def, blast)
|
|
160 |
|
|
161 |
(*Monotonicity*)
|
|
162 |
lemma keysFor_mono: "G\<subseteq>H ==> keysFor(G) \<subseteq> keysFor(H)"
|
|
163 |
by (unfold keysFor_def, blast)
|
|
164 |
|
|
165 |
lemma keysFor_insert_Agent [simp]: "keysFor (insert (Agent A) H) = keysFor H"
|
|
166 |
by (unfold keysFor_def, auto)
|
|
167 |
|
|
168 |
lemma keysFor_insert_Nonce [simp]: "keysFor (insert (Nonce N) H) = keysFor H"
|
|
169 |
by (unfold keysFor_def, auto)
|
|
170 |
|
|
171 |
lemma keysFor_insert_Number [simp]: "keysFor (insert (Number N) H) = keysFor H"
|
|
172 |
by (unfold keysFor_def, auto)
|
|
173 |
|
|
174 |
lemma keysFor_insert_Key [simp]: "keysFor (insert (Key K) H) = keysFor H"
|
|
175 |
by (unfold keysFor_def, auto)
|
|
176 |
|
|
177 |
lemma keysFor_insert_Pan [simp]: "keysFor (insert (Pan A) H) = keysFor H"
|
|
178 |
by (unfold keysFor_def, auto)
|
|
179 |
|
|
180 |
lemma keysFor_insert_Hash [simp]: "keysFor (insert (Hash X) H) = keysFor H"
|
|
181 |
by (unfold keysFor_def, auto)
|
|
182 |
|
|
183 |
lemma keysFor_insert_MPair [simp]: "keysFor (insert {|X,Y|} H) = keysFor H"
|
|
184 |
by (unfold keysFor_def, auto)
|
|
185 |
|
|
186 |
lemma keysFor_insert_Crypt [simp]:
|
|
187 |
"keysFor (insert (Crypt K X) H) = insert (invKey K) (keysFor H)"
|
|
188 |
by (unfold keysFor_def, auto)
|
|
189 |
|
|
190 |
lemma keysFor_image_Key [simp]: "keysFor (Key`E) = {}"
|
|
191 |
by (unfold keysFor_def, auto)
|
|
192 |
|
|
193 |
lemma Crypt_imp_invKey_keysFor: "Crypt K X \<in> H ==> invKey K \<in> keysFor H"
|
|
194 |
by (unfold keysFor_def, blast)
|
|
195 |
|
|
196 |
|
|
197 |
subsection{*Inductive relation "parts"*}
|
|
198 |
|
|
199 |
lemma MPair_parts:
|
|
200 |
"[| {|X,Y|} \<in> parts H;
|
|
201 |
[| X \<in> parts H; Y \<in> parts H |] ==> P |] ==> P"
|
|
202 |
by (blast dest: parts.Fst parts.Snd)
|
|
203 |
|
|
204 |
declare MPair_parts [elim!] parts.Body [dest!]
|
|
205 |
text{*NB These two rules are UNSAFE in the formal sense, as they discard the
|
|
206 |
compound message. They work well on THIS FILE.
|
|
207 |
@{text MPair_parts} is left as SAFE because it speeds up proofs.
|
|
208 |
The Crypt rule is normally kept UNSAFE to avoid breaking up certificates.*}
|
|
209 |
|
|
210 |
lemma parts_increasing: "H \<subseteq> parts(H)"
|
|
211 |
by blast
|
|
212 |
|
|
213 |
lemmas parts_insertI = subset_insertI [THEN parts_mono, THEN subsetD, standard]
|
|
214 |
|
|
215 |
lemma parts_empty [simp]: "parts{} = {}"
|
|
216 |
apply safe
|
|
217 |
apply (erule parts.induct, blast+)
|
|
218 |
done
|
|
219 |
|
|
220 |
lemma parts_emptyE [elim!]: "X\<in> parts{} ==> P"
|
|
221 |
by simp
|
|
222 |
|
|
223 |
(*WARNING: loops if H = {Y}, therefore must not be repeated!*)
|
|
224 |
lemma parts_singleton: "X\<in> parts H ==> \<exists>Y\<in>H. X\<in> parts {Y}"
|
|
225 |
by (erule parts.induct, blast+)
|
|
226 |
|
|
227 |
|
|
228 |
subsubsection{*Unions*}
|
|
229 |
|
|
230 |
lemma parts_Un_subset1: "parts(G) \<union> parts(H) \<subseteq> parts(G \<union> H)"
|
|
231 |
by (intro Un_least parts_mono Un_upper1 Un_upper2)
|
|
232 |
|
|
233 |
lemma parts_Un_subset2: "parts(G \<union> H) \<subseteq> parts(G) \<union> parts(H)"
|
|
234 |
apply (rule subsetI)
|
|
235 |
apply (erule parts.induct, blast+)
|
|
236 |
done
|
|
237 |
|
|
238 |
lemma parts_Un [simp]: "parts(G \<union> H) = parts(G) \<union> parts(H)"
|
|
239 |
by (intro equalityI parts_Un_subset1 parts_Un_subset2)
|
|
240 |
|
|
241 |
lemma parts_insert: "parts (insert X H) = parts {X} \<union> parts H"
|
|
242 |
apply (subst insert_is_Un [of _ H])
|
|
243 |
apply (simp only: parts_Un)
|
|
244 |
done
|
|
245 |
|
|
246 |
(*TWO inserts to avoid looping. This rewrite is better than nothing.
|
|
247 |
Not suitable for Addsimps: its behaviour can be strange.*)
|
|
248 |
lemma parts_insert2:
|
|
249 |
"parts (insert X (insert Y H)) = parts {X} \<union> parts {Y} \<union> parts H"
|
|
250 |
apply (simp add: Un_assoc)
|
|
251 |
apply (simp add: parts_insert [symmetric])
|
|
252 |
done
|
|
253 |
|
|
254 |
lemma parts_UN_subset1: "(\<Union>x\<in>A. parts(H x)) \<subseteq> parts(\<Union>x\<in>A. H x)"
|
|
255 |
by (intro UN_least parts_mono UN_upper)
|
|
256 |
|
|
257 |
lemma parts_UN_subset2: "parts(\<Union>x\<in>A. H x) \<subseteq> (\<Union>x\<in>A. parts(H x))"
|
|
258 |
apply (rule subsetI)
|
|
259 |
apply (erule parts.induct, blast+)
|
|
260 |
done
|
|
261 |
|
|
262 |
lemma parts_UN [simp]: "parts(\<Union>x\<in>A. H x) = (\<Union>x\<in>A. parts(H x))"
|
|
263 |
by (intro equalityI parts_UN_subset1 parts_UN_subset2)
|
|
264 |
|
|
265 |
(*Added to simplify arguments to parts, analz and synth.
|
|
266 |
NOTE: the UN versions are no longer used!*)
|
|
267 |
|
|
268 |
|
|
269 |
text{*This allows @{text blast} to simplify occurrences of
|
|
270 |
@{term "parts(G\<union>H)"} in the assumption.*}
|
|
271 |
declare parts_Un [THEN equalityD1, THEN subsetD, THEN UnE, elim!]
|
|
272 |
|
|
273 |
|
|
274 |
lemma parts_insert_subset: "insert X (parts H) \<subseteq> parts(insert X H)"
|
|
275 |
by (blast intro: parts_mono [THEN [2] rev_subsetD])
|
|
276 |
|
|
277 |
subsubsection{*Idempotence and transitivity*}
|
|
278 |
|
|
279 |
lemma parts_partsD [dest!]: "X\<in> parts (parts H) ==> X\<in> parts H"
|
|
280 |
by (erule parts.induct, blast+)
|
|
281 |
|
|
282 |
lemma parts_idem [simp]: "parts (parts H) = parts H"
|
|
283 |
by blast
|
|
284 |
|
|
285 |
lemma parts_trans: "[| X\<in> parts G; G \<subseteq> parts H |] ==> X\<in> parts H"
|
|
286 |
by (drule parts_mono, blast)
|
|
287 |
|
|
288 |
(*Cut*)
|
|
289 |
lemma parts_cut:
|
|
290 |
"[| Y\<in> parts (insert X G); X\<in> parts H |] ==> Y\<in> parts (G \<union> H)"
|
|
291 |
by (erule parts_trans, auto)
|
|
292 |
|
|
293 |
lemma parts_cut_eq [simp]: "X\<in> parts H ==> parts (insert X H) = parts H"
|
|
294 |
by (force dest!: parts_cut intro: parts_insertI)
|
|
295 |
|
|
296 |
|
|
297 |
subsubsection{*Rewrite rules for pulling out atomic messages*}
|
|
298 |
|
|
299 |
lemmas parts_insert_eq_I = equalityI [OF subsetI parts_insert_subset]
|
|
300 |
|
|
301 |
|
|
302 |
lemma parts_insert_Agent [simp]:
|
|
303 |
"parts (insert (Agent agt) H) = insert (Agent agt) (parts H)"
|
|
304 |
apply (rule parts_insert_eq_I)
|
|
305 |
apply (erule parts.induct, auto)
|
|
306 |
done
|
|
307 |
|
|
308 |
lemma parts_insert_Nonce [simp]:
|
|
309 |
"parts (insert (Nonce N) H) = insert (Nonce N) (parts H)"
|
|
310 |
apply (rule parts_insert_eq_I)
|
|
311 |
apply (erule parts.induct, auto)
|
|
312 |
done
|
|
313 |
|
|
314 |
lemma parts_insert_Number [simp]:
|
|
315 |
"parts (insert (Number N) H) = insert (Number N) (parts H)"
|
|
316 |
apply (rule parts_insert_eq_I)
|
|
317 |
apply (erule parts.induct, auto)
|
|
318 |
done
|
|
319 |
|
|
320 |
lemma parts_insert_Key [simp]:
|
|
321 |
"parts (insert (Key K) H) = insert (Key K) (parts H)"
|
|
322 |
apply (rule parts_insert_eq_I)
|
|
323 |
apply (erule parts.induct, auto)
|
|
324 |
done
|
|
325 |
|
|
326 |
lemma parts_insert_Pan [simp]:
|
|
327 |
"parts (insert (Pan A) H) = insert (Pan A) (parts H)"
|
|
328 |
apply (rule parts_insert_eq_I)
|
|
329 |
apply (erule parts.induct, auto)
|
|
330 |
done
|
|
331 |
|
|
332 |
lemma parts_insert_Hash [simp]:
|
|
333 |
"parts (insert (Hash X) H) = insert (Hash X) (parts H)"
|
|
334 |
apply (rule parts_insert_eq_I)
|
|
335 |
apply (erule parts.induct, auto)
|
|
336 |
done
|
|
337 |
|
|
338 |
lemma parts_insert_Crypt [simp]:
|
|
339 |
"parts (insert (Crypt K X) H) =
|
|
340 |
insert (Crypt K X) (parts (insert X H))"
|
|
341 |
apply (rule equalityI)
|
|
342 |
apply (rule subsetI)
|
|
343 |
apply (erule parts.induct, auto)
|
|
344 |
apply (erule parts.induct)
|
|
345 |
apply (blast intro: parts.Body)+
|
|
346 |
done
|
|
347 |
|
|
348 |
lemma parts_insert_MPair [simp]:
|
|
349 |
"parts (insert {|X,Y|} H) =
|
|
350 |
insert {|X,Y|} (parts (insert X (insert Y H)))"
|
|
351 |
apply (rule equalityI)
|
|
352 |
apply (rule subsetI)
|
|
353 |
apply (erule parts.induct, auto)
|
|
354 |
apply (erule parts.induct)
|
|
355 |
apply (blast intro: parts.Fst parts.Snd)+
|
|
356 |
done
|
|
357 |
|
|
358 |
lemma parts_image_Key [simp]: "parts (Key`N) = Key`N"
|
|
359 |
apply auto
|
|
360 |
apply (erule parts.induct, auto)
|
|
361 |
done
|
|
362 |
|
|
363 |
lemma parts_image_Pan [simp]: "parts (Pan`A) = Pan`A"
|
|
364 |
apply auto
|
|
365 |
apply (erule parts.induct, auto)
|
|
366 |
done
|
|
367 |
|
|
368 |
|
|
369 |
(*In any message, there is an upper bound N on its greatest nonce.*)
|
|
370 |
lemma msg_Nonce_supply: "\<exists>N. \<forall>n. N\<le>n --> Nonce n \<notin> parts {msg}"
|
|
371 |
apply (induct_tac "msg")
|
|
372 |
apply (simp_all (no_asm_simp) add: exI parts_insert2)
|
|
373 |
(*MPair case: blast_tac works out the necessary sum itself!*)
|
|
374 |
prefer 2 apply (blast elim!: add_leE)
|
|
375 |
(*Nonce case*)
|
|
376 |
apply (rule_tac x = "N + Suc nat" in exI)
|
|
377 |
apply (auto elim!: add_leE)
|
|
378 |
done
|
|
379 |
|
|
380 |
(* Ditto, for numbers.*)
|
|
381 |
lemma msg_Number_supply: "\<exists>N. \<forall>n. N<=n --> Number n \<notin> parts {msg}"
|
|
382 |
apply (induct_tac "msg")
|
|
383 |
apply (simp_all (no_asm_simp) add: exI parts_insert2)
|
|
384 |
prefer 2 apply (blast elim!: add_leE)
|
|
385 |
apply (rule_tac x = "N + Suc nat" in exI, auto)
|
|
386 |
done
|
|
387 |
|
|
388 |
subsection{*Inductive relation "analz"*}
|
|
389 |
|
|
390 |
text{*Inductive definition of "analz" -- what can be broken down from a set of
|
|
391 |
messages, including keys. A form of downward closure. Pairs can
|
|
392 |
be taken apart; messages decrypted with known keys.*}
|
|
393 |
|
|
394 |
consts analz :: "msg set => msg set"
|
|
395 |
inductive "analz H"
|
|
396 |
intros
|
|
397 |
Inj [intro,simp] : "X \<in> H ==> X \<in> analz H"
|
|
398 |
Fst: "{|X,Y|} \<in> analz H ==> X \<in> analz H"
|
|
399 |
Snd: "{|X,Y|} \<in> analz H ==> Y \<in> analz H"
|
|
400 |
Decrypt [dest]:
|
|
401 |
"[|Crypt K X \<in> analz H; Key(invKey K): analz H|] ==> X \<in> analz H"
|
|
402 |
|
|
403 |
|
|
404 |
(*Monotonicity; Lemma 1 of Lowe's paper*)
|
|
405 |
lemma analz_mono: "G<=H ==> analz(G) <= analz(H)"
|
|
406 |
apply auto
|
|
407 |
apply (erule analz.induct)
|
|
408 |
apply (auto dest: Fst Snd)
|
|
409 |
done
|
|
410 |
|
|
411 |
text{*Making it safe speeds up proofs*}
|
|
412 |
lemma MPair_analz [elim!]:
|
|
413 |
"[| {|X,Y|} \<in> analz H;
|
|
414 |
[| X \<in> analz H; Y \<in> analz H |] ==> P
|
|
415 |
|] ==> P"
|
|
416 |
by (blast dest: analz.Fst analz.Snd)
|
|
417 |
|
|
418 |
lemma analz_increasing: "H \<subseteq> analz(H)"
|
|
419 |
by blast
|
|
420 |
|
|
421 |
lemma analz_subset_parts: "analz H \<subseteq> parts H"
|
|
422 |
apply (rule subsetI)
|
|
423 |
apply (erule analz.induct, blast+)
|
|
424 |
done
|
|
425 |
|
|
426 |
lemmas analz_into_parts = analz_subset_parts [THEN subsetD, standard]
|
|
427 |
|
|
428 |
lemmas not_parts_not_analz = analz_subset_parts [THEN contra_subsetD, standard]
|
|
429 |
|
|
430 |
|
|
431 |
lemma parts_analz [simp]: "parts (analz H) = parts H"
|
|
432 |
apply (rule equalityI)
|
|
433 |
apply (rule analz_subset_parts [THEN parts_mono, THEN subset_trans], simp)
|
|
434 |
apply (blast intro: analz_increasing [THEN parts_mono, THEN subsetD])
|
|
435 |
done
|
|
436 |
|
|
437 |
lemma analz_parts [simp]: "analz (parts H) = parts H"
|
|
438 |
apply auto
|
|
439 |
apply (erule analz.induct, auto)
|
|
440 |
done
|
|
441 |
|
|
442 |
lemmas analz_insertI = subset_insertI [THEN analz_mono, THEN [2] rev_subsetD, standard]
|
|
443 |
|
|
444 |
subsubsection{*General equational properties*}
|
|
445 |
|
|
446 |
lemma analz_empty [simp]: "analz{} = {}"
|
|
447 |
apply safe
|
|
448 |
apply (erule analz.induct, blast+)
|
|
449 |
done
|
|
450 |
|
|
451 |
(*Converse fails: we can analz more from the union than from the
|
|
452 |
separate parts, as a key in one might decrypt a message in the other*)
|
|
453 |
lemma analz_Un: "analz(G) \<union> analz(H) \<subseteq> analz(G \<union> H)"
|
|
454 |
by (intro Un_least analz_mono Un_upper1 Un_upper2)
|
|
455 |
|
|
456 |
lemma analz_insert: "insert X (analz H) \<subseteq> analz(insert X H)"
|
|
457 |
by (blast intro: analz_mono [THEN [2] rev_subsetD])
|
|
458 |
|
|
459 |
subsubsection{*Rewrite rules for pulling out atomic messages*}
|
|
460 |
|
|
461 |
lemmas analz_insert_eq_I = equalityI [OF subsetI analz_insert]
|
|
462 |
|
|
463 |
lemma analz_insert_Agent [simp]:
|
|
464 |
"analz (insert (Agent agt) H) = insert (Agent agt) (analz H)"
|
|
465 |
apply (rule analz_insert_eq_I)
|
|
466 |
apply (erule analz.induct, auto)
|
|
467 |
done
|
|
468 |
|
|
469 |
lemma analz_insert_Nonce [simp]:
|
|
470 |
"analz (insert (Nonce N) H) = insert (Nonce N) (analz H)"
|
|
471 |
apply (rule analz_insert_eq_I)
|
|
472 |
apply (erule analz.induct, auto)
|
|
473 |
done
|
|
474 |
|
|
475 |
lemma analz_insert_Number [simp]:
|
|
476 |
"analz (insert (Number N) H) = insert (Number N) (analz H)"
|
|
477 |
apply (rule analz_insert_eq_I)
|
|
478 |
apply (erule analz.induct, auto)
|
|
479 |
done
|
|
480 |
|
|
481 |
lemma analz_insert_Hash [simp]:
|
|
482 |
"analz (insert (Hash X) H) = insert (Hash X) (analz H)"
|
|
483 |
apply (rule analz_insert_eq_I)
|
|
484 |
apply (erule analz.induct, auto)
|
|
485 |
done
|
|
486 |
|
|
487 |
(*Can only pull out Keys if they are not needed to decrypt the rest*)
|
|
488 |
lemma analz_insert_Key [simp]:
|
|
489 |
"K \<notin> keysFor (analz H) ==>
|
|
490 |
analz (insert (Key K) H) = insert (Key K) (analz H)"
|
|
491 |
apply (unfold keysFor_def)
|
|
492 |
apply (rule analz_insert_eq_I)
|
|
493 |
apply (erule analz.induct, auto)
|
|
494 |
done
|
|
495 |
|
|
496 |
lemma analz_insert_MPair [simp]:
|
|
497 |
"analz (insert {|X,Y|} H) =
|
|
498 |
insert {|X,Y|} (analz (insert X (insert Y H)))"
|
|
499 |
apply (rule equalityI)
|
|
500 |
apply (rule subsetI)
|
|
501 |
apply (erule analz.induct, auto)
|
|
502 |
apply (erule analz.induct)
|
|
503 |
apply (blast intro: analz.Fst analz.Snd)+
|
|
504 |
done
|
|
505 |
|
|
506 |
(*Can pull out enCrypted message if the Key is not known*)
|
|
507 |
lemma analz_insert_Crypt:
|
|
508 |
"Key (invKey K) \<notin> analz H
|
|
509 |
==> analz (insert (Crypt K X) H) = insert (Crypt K X) (analz H)"
|
|
510 |
apply (rule analz_insert_eq_I)
|
|
511 |
apply (erule analz.induct, auto)
|
|
512 |
done
|
|
513 |
|
|
514 |
lemma analz_insert_Pan [simp]:
|
|
515 |
"analz (insert (Pan A) H) = insert (Pan A) (analz H)"
|
|
516 |
apply (rule analz_insert_eq_I)
|
|
517 |
apply (erule analz.induct, auto)
|
|
518 |
done
|
|
519 |
|
|
520 |
lemma lemma1: "Key (invKey K) \<in> analz H ==>
|
|
521 |
analz (insert (Crypt K X) H) \<subseteq>
|
|
522 |
insert (Crypt K X) (analz (insert X H))"
|
|
523 |
apply (rule subsetI)
|
|
524 |
apply (erule_tac xa = x in analz.induct, auto)
|
|
525 |
done
|
|
526 |
|
|
527 |
lemma lemma2: "Key (invKey K) \<in> analz H ==>
|
|
528 |
insert (Crypt K X) (analz (insert X H)) \<subseteq>
|
|
529 |
analz (insert (Crypt K X) H)"
|
|
530 |
apply auto
|
|
531 |
apply (erule_tac xa = x in analz.induct, auto)
|
|
532 |
apply (blast intro: analz_insertI analz.Decrypt)
|
|
533 |
done
|
|
534 |
|
|
535 |
lemma analz_insert_Decrypt:
|
|
536 |
"Key (invKey K) \<in> analz H ==>
|
|
537 |
analz (insert (Crypt K X) H) =
|
|
538 |
insert (Crypt K X) (analz (insert X H))"
|
|
539 |
by (intro equalityI lemma1 lemma2)
|
|
540 |
|
|
541 |
(*Case analysis: either the message is secure, or it is not!
|
|
542 |
Effective, but can cause subgoals to blow up!
|
|
543 |
Use with split_if; apparently split_tac does not cope with patterns
|
|
544 |
such as "analz (insert (Crypt K X) H)" *)
|
|
545 |
lemma analz_Crypt_if [simp]:
|
|
546 |
"analz (insert (Crypt K X) H) =
|
|
547 |
(if (Key (invKey K) \<in> analz H)
|
|
548 |
then insert (Crypt K X) (analz (insert X H))
|
|
549 |
else insert (Crypt K X) (analz H))"
|
|
550 |
by (simp add: analz_insert_Crypt analz_insert_Decrypt)
|
|
551 |
|
|
552 |
|
|
553 |
(*This rule supposes "for the sake of argument" that we have the key.*)
|
|
554 |
lemma analz_insert_Crypt_subset:
|
|
555 |
"analz (insert (Crypt K X) H) \<subseteq>
|
|
556 |
insert (Crypt K X) (analz (insert X H))"
|
|
557 |
apply (rule subsetI)
|
|
558 |
apply (erule analz.induct, auto)
|
|
559 |
done
|
|
560 |
|
|
561 |
lemma analz_image_Key [simp]: "analz (Key`N) = Key`N"
|
|
562 |
apply auto
|
|
563 |
apply (erule analz.induct, auto)
|
|
564 |
done
|
|
565 |
|
|
566 |
lemma analz_image_Pan [simp]: "analz (Pan`A) = Pan`A"
|
|
567 |
apply auto
|
|
568 |
apply (erule analz.induct, auto)
|
|
569 |
done
|
|
570 |
|
|
571 |
|
|
572 |
subsubsection{*Idempotence and transitivity*}
|
|
573 |
|
|
574 |
lemma analz_analzD [dest!]: "X\<in> analz (analz H) ==> X\<in> analz H"
|
|
575 |
by (erule analz.induct, blast+)
|
|
576 |
|
|
577 |
lemma analz_idem [simp]: "analz (analz H) = analz H"
|
|
578 |
by blast
|
|
579 |
|
|
580 |
lemma analz_trans: "[| X\<in> analz G; G \<subseteq> analz H |] ==> X\<in> analz H"
|
|
581 |
by (drule analz_mono, blast)
|
|
582 |
|
|
583 |
(*Cut; Lemma 2 of Lowe*)
|
|
584 |
lemma analz_cut: "[| Y\<in> analz (insert X H); X\<in> analz H |] ==> Y\<in> analz H"
|
|
585 |
by (erule analz_trans, blast)
|
|
586 |
|
|
587 |
(*Cut can be proved easily by induction on
|
|
588 |
"Y: analz (insert X H) ==> X: analz H --> Y: analz H"
|
|
589 |
*)
|
|
590 |
|
|
591 |
(*This rewrite rule helps in the simplification of messages that involve
|
|
592 |
the forwarding of unknown components (X). Without it, removing occurrences
|
|
593 |
of X can be very complicated. *)
|
|
594 |
lemma analz_insert_eq: "X\<in> analz H ==> analz (insert X H) = analz H"
|
|
595 |
by (blast intro: analz_cut analz_insertI)
|
|
596 |
|
|
597 |
|
|
598 |
text{*A congruence rule for "analz"*}
|
|
599 |
|
|
600 |
lemma analz_subset_cong:
|
|
601 |
"[| analz G \<subseteq> analz G'; analz H \<subseteq> analz H'
|
|
602 |
|] ==> analz (G \<union> H) \<subseteq> analz (G' \<union> H')"
|
|
603 |
apply clarify
|
|
604 |
apply (erule analz.induct)
|
|
605 |
apply (best intro: analz_mono [THEN subsetD])+
|
|
606 |
done
|
|
607 |
|
|
608 |
lemma analz_cong:
|
|
609 |
"[| analz G = analz G'; analz H = analz H'
|
|
610 |
|] ==> analz (G \<union> H) = analz (G' \<union> H')"
|
|
611 |
by (intro equalityI analz_subset_cong, simp_all)
|
|
612 |
|
|
613 |
lemma analz_insert_cong:
|
|
614 |
"analz H = analz H' ==> analz(insert X H) = analz(insert X H')"
|
|
615 |
by (force simp only: insert_def intro!: analz_cong)
|
|
616 |
|
|
617 |
(*If there are no pairs or encryptions then analz does nothing*)
|
|
618 |
lemma analz_trivial:
|
|
619 |
"[| \<forall>X Y. {|X,Y|} \<notin> H; \<forall>X K. Crypt K X \<notin> H |] ==> analz H = H"
|
|
620 |
apply safe
|
|
621 |
apply (erule analz.induct, blast+)
|
|
622 |
done
|
|
623 |
|
|
624 |
(*These two are obsolete (with a single Spy) but cost little to prove...*)
|
|
625 |
lemma analz_UN_analz_lemma:
|
|
626 |
"X\<in> analz (\<Union>i\<in>A. analz (H i)) ==> X\<in> analz (\<Union>i\<in>A. H i)"
|
|
627 |
apply (erule analz.induct)
|
|
628 |
apply (blast intro: analz_mono [THEN [2] rev_subsetD])+
|
|
629 |
done
|
|
630 |
|
|
631 |
lemma analz_UN_analz [simp]: "analz (\<Union>i\<in>A. analz (H i)) = analz (\<Union>i\<in>A. H i)"
|
|
632 |
by (blast intro: analz_UN_analz_lemma analz_mono [THEN [2] rev_subsetD])
|
|
633 |
|
|
634 |
|
|
635 |
subsection{*Inductive relation "synth"*}
|
|
636 |
|
|
637 |
text{*Inductive definition of "synth" -- what can be built up from a set of
|
|
638 |
messages. A form of upward closure. Pairs can be built, messages
|
|
639 |
encrypted with known keys. Agent names are public domain.
|
|
640 |
Numbers can be guessed, but Nonces cannot be.*}
|
|
641 |
|
|
642 |
consts synth :: "msg set => msg set"
|
|
643 |
inductive "synth H"
|
|
644 |
intros
|
|
645 |
Inj [intro]: "X \<in> H ==> X \<in> synth H"
|
|
646 |
Agent [intro]: "Agent agt \<in> synth H"
|
|
647 |
Number [intro]: "Number n \<in> synth H"
|
|
648 |
Hash [intro]: "X \<in> synth H ==> Hash X \<in> synth H"
|
|
649 |
MPair [intro]: "[|X \<in> synth H; Y \<in> synth H|] ==> {|X,Y|} \<in> synth H"
|
|
650 |
Crypt [intro]: "[|X \<in> synth H; Key(K) \<in> H|] ==> Crypt K X \<in> synth H"
|
|
651 |
|
|
652 |
(*Monotonicity*)
|
|
653 |
lemma synth_mono: "G<=H ==> synth(G) <= synth(H)"
|
|
654 |
apply auto
|
|
655 |
apply (erule synth.induct)
|
|
656 |
apply (auto dest: Fst Snd Body)
|
|
657 |
done
|
|
658 |
|
|
659 |
(*NO Agent_synth, as any Agent name can be synthesized. Ditto for Number*)
|
|
660 |
inductive_cases Nonce_synth [elim!]: "Nonce n \<in> synth H"
|
|
661 |
inductive_cases Key_synth [elim!]: "Key K \<in> synth H"
|
|
662 |
inductive_cases Hash_synth [elim!]: "Hash X \<in> synth H"
|
|
663 |
inductive_cases MPair_synth [elim!]: "{|X,Y|} \<in> synth H"
|
|
664 |
inductive_cases Crypt_synth [elim!]: "Crypt K X \<in> synth H"
|
|
665 |
inductive_cases Pan_synth [elim!]: "Pan A \<in> synth H"
|
|
666 |
|
|
667 |
|
|
668 |
lemma synth_increasing: "H \<subseteq> synth(H)"
|
|
669 |
by blast
|
|
670 |
|
|
671 |
subsubsection{*Unions*}
|
|
672 |
|
|
673 |
(*Converse fails: we can synth more from the union than from the
|
|
674 |
separate parts, building a compound message using elements of each.*)
|
|
675 |
lemma synth_Un: "synth(G) \<union> synth(H) \<subseteq> synth(G \<union> H)"
|
|
676 |
by (intro Un_least synth_mono Un_upper1 Un_upper2)
|
|
677 |
|
|
678 |
lemma synth_insert: "insert X (synth H) \<subseteq> synth(insert X H)"
|
|
679 |
by (blast intro: synth_mono [THEN [2] rev_subsetD])
|
|
680 |
|
|
681 |
subsubsection{*Idempotence and transitivity*}
|
|
682 |
|
|
683 |
lemma synth_synthD [dest!]: "X\<in> synth (synth H) ==> X\<in> synth H"
|
|
684 |
by (erule synth.induct, blast+)
|
|
685 |
|
|
686 |
lemma synth_idem: "synth (synth H) = synth H"
|
|
687 |
by blast
|
|
688 |
|
|
689 |
lemma synth_trans: "[| X\<in> synth G; G \<subseteq> synth H |] ==> X\<in> synth H"
|
|
690 |
by (drule synth_mono, blast)
|
|
691 |
|
|
692 |
(*Cut; Lemma 2 of Lowe*)
|
|
693 |
lemma synth_cut: "[| Y\<in> synth (insert X H); X\<in> synth H |] ==> Y\<in> synth H"
|
|
694 |
by (erule synth_trans, blast)
|
|
695 |
|
|
696 |
lemma Agent_synth [simp]: "Agent A \<in> synth H"
|
|
697 |
by blast
|
|
698 |
|
|
699 |
lemma Number_synth [simp]: "Number n \<in> synth H"
|
|
700 |
by blast
|
|
701 |
|
|
702 |
lemma Nonce_synth_eq [simp]: "(Nonce N \<in> synth H) = (Nonce N \<in> H)"
|
|
703 |
by blast
|
|
704 |
|
|
705 |
lemma Key_synth_eq [simp]: "(Key K \<in> synth H) = (Key K \<in> H)"
|
|
706 |
by blast
|
|
707 |
|
|
708 |
lemma Crypt_synth_eq [simp]: "Key K \<notin> H ==> (Crypt K X \<in> synth H) = (Crypt K X \<in> H)"
|
|
709 |
by blast
|
|
710 |
|
|
711 |
lemma Pan_synth_eq [simp]: "(Pan A \<in> synth H) = (Pan A \<in> H)"
|
|
712 |
by blast
|
|
713 |
|
|
714 |
lemma keysFor_synth [simp]:
|
|
715 |
"keysFor (synth H) = keysFor H \<union> invKey`{K. Key K \<in> H}"
|
|
716 |
by (unfold keysFor_def, blast)
|
|
717 |
|
|
718 |
|
|
719 |
subsubsection{*Combinations of parts, analz and synth*}
|
|
720 |
|
|
721 |
lemma parts_synth [simp]: "parts (synth H) = parts H \<union> synth H"
|
|
722 |
apply (rule equalityI)
|
|
723 |
apply (rule subsetI)
|
|
724 |
apply (erule parts.induct)
|
|
725 |
apply (blast intro: synth_increasing [THEN parts_mono, THEN subsetD]
|
|
726 |
parts.Fst parts.Snd parts.Body)+
|
|
727 |
done
|
|
728 |
|
|
729 |
lemma analz_analz_Un [simp]: "analz (analz G \<union> H) = analz (G \<union> H)"
|
|
730 |
apply (intro equalityI analz_subset_cong)+
|
|
731 |
apply simp_all
|
|
732 |
done
|
|
733 |
|
|
734 |
lemma analz_synth_Un [simp]: "analz (synth G \<union> H) = analz (G \<union> H) \<union> synth G"
|
|
735 |
apply (rule equalityI)
|
|
736 |
apply (rule subsetI)
|
|
737 |
apply (erule analz.induct)
|
|
738 |
prefer 5 apply (blast intro: analz_mono [THEN [2] rev_subsetD])
|
|
739 |
apply (blast intro: analz.Fst analz.Snd analz.Decrypt)+
|
|
740 |
done
|
|
741 |
|
|
742 |
lemma analz_synth [simp]: "analz (synth H) = analz H \<union> synth H"
|
|
743 |
apply (cut_tac H = "{}" in analz_synth_Un)
|
|
744 |
apply (simp (no_asm_use))
|
|
745 |
done
|
|
746 |
|
|
747 |
|
|
748 |
subsubsection{*For reasoning about the Fake rule in traces*}
|
|
749 |
|
|
750 |
lemma parts_insert_subset_Un: "X\<in> G ==> parts(insert X H) \<subseteq> parts G \<union> parts H"
|
|
751 |
by (rule subset_trans [OF parts_mono parts_Un_subset2], blast)
|
|
752 |
|
|
753 |
(*More specifically for Fake. Very occasionally we could do with a version
|
|
754 |
of the form parts{X} \<subseteq> synth (analz H) \<union> parts H *)
|
|
755 |
lemma Fake_parts_insert: "X \<in> synth (analz H) ==>
|
|
756 |
parts (insert X H) \<subseteq> synth (analz H) \<union> parts H"
|
|
757 |
apply (drule parts_insert_subset_Un)
|
|
758 |
apply (simp (no_asm_use))
|
|
759 |
apply blast
|
|
760 |
done
|
|
761 |
|
|
762 |
lemma Fake_parts_insert_in_Un:
|
|
763 |
"[|Z \<in> parts (insert X H); X: synth (analz H)|]
|
|
764 |
==> Z \<in> synth (analz H) \<union> parts H";
|
|
765 |
by (blast dest: Fake_parts_insert [THEN subsetD, dest])
|
|
766 |
|
|
767 |
(*H is sometimes (Key ` KK \<union> spies evs), so can't put G=H*)
|
|
768 |
lemma Fake_analz_insert:
|
|
769 |
"X\<in> synth (analz G) ==>
|
|
770 |
analz (insert X H) \<subseteq> synth (analz G) \<union> analz (G \<union> H)"
|
|
771 |
apply (rule subsetI)
|
|
772 |
apply (subgoal_tac "x \<in> analz (synth (analz G) \<union> H) ")
|
|
773 |
prefer 2 apply (blast intro: analz_mono [THEN [2] rev_subsetD] analz_mono [THEN synth_mono, THEN [2] rev_subsetD])
|
|
774 |
apply (simp (no_asm_use))
|
|
775 |
apply blast
|
|
776 |
done
|
|
777 |
|
|
778 |
lemma analz_conj_parts [simp]:
|
|
779 |
"(X \<in> analz H & X \<in> parts H) = (X \<in> analz H)"
|
|
780 |
by (blast intro: analz_subset_parts [THEN subsetD])
|
|
781 |
|
|
782 |
lemma analz_disj_parts [simp]:
|
|
783 |
"(X \<in> analz H | X \<in> parts H) = (X \<in> parts H)"
|
|
784 |
by (blast intro: analz_subset_parts [THEN subsetD])
|
|
785 |
|
|
786 |
(*Without this equation, other rules for synth and analz would yield
|
|
787 |
redundant cases*)
|
|
788 |
lemma MPair_synth_analz [iff]:
|
|
789 |
"({|X,Y|} \<in> synth (analz H)) =
|
|
790 |
(X \<in> synth (analz H) & Y \<in> synth (analz H))"
|
|
791 |
by blast
|
|
792 |
|
|
793 |
lemma Crypt_synth_analz:
|
|
794 |
"[| Key K \<in> analz H; Key (invKey K) \<in> analz H |]
|
|
795 |
==> (Crypt K X \<in> synth (analz H)) = (X \<in> synth (analz H))"
|
|
796 |
by blast
|
|
797 |
|
|
798 |
|
|
799 |
lemma Hash_synth_analz [simp]:
|
|
800 |
"X \<notin> synth (analz H)
|
|
801 |
==> (Hash{|X,Y|} \<in> synth (analz H)) = (Hash{|X,Y|} \<in> analz H)"
|
|
802 |
by blast
|
|
803 |
|
|
804 |
|
|
805 |
(*We do NOT want Crypt... messages broken up in protocols!!*)
|
|
806 |
declare parts.Body [rule del]
|
|
807 |
|
|
808 |
|
|
809 |
text{*Rewrites to push in Key and Crypt messages, so that other messages can
|
|
810 |
be pulled out using the @{text analz_insert} rules*}
|
|
811 |
ML
|
|
812 |
{*
|
|
813 |
fun insComm x y = inst "x" x (inst "y" y insert_commute);
|
|
814 |
|
|
815 |
bind_thms ("pushKeys",
|
|
816 |
map (insComm "Key ?K")
|
|
817 |
["Agent ?C", "Nonce ?N", "Number ?N", "Pan ?PAN",
|
|
818 |
"Hash ?X", "MPair ?X ?Y", "Crypt ?X ?K'"]);
|
|
819 |
|
|
820 |
bind_thms ("pushCrypts",
|
|
821 |
map (insComm "Crypt ?X ?K")
|
|
822 |
["Agent ?C", "Nonce ?N", "Number ?N", "Pan ?PAN",
|
|
823 |
"Hash ?X'", "MPair ?X' ?Y"]);
|
|
824 |
*}
|
|
825 |
|
|
826 |
text{*Cannot be added with @{text "[simp]"} -- messages should not always be
|
|
827 |
re-ordered.*}
|
|
828 |
lemmas pushes = pushKeys pushCrypts
|
|
829 |
|
|
830 |
|
|
831 |
subsection{*Tactics useful for many protocol proofs*}
|
14218
|
832 |
(*<*)
|
14199
|
833 |
ML
|
|
834 |
{*
|
|
835 |
val analz_increasing = thm "analz_increasing";
|
|
836 |
val analz_subset_parts = thm "analz_subset_parts";
|
|
837 |
val parts_analz = thm "parts_analz";
|
|
838 |
val analz_parts = thm "analz_parts";
|
|
839 |
val analz_insertI = thm "analz_insertI";
|
|
840 |
val Fake_parts_insert = thm "Fake_parts_insert";
|
|
841 |
val Fake_analz_insert = thm "Fake_analz_insert";
|
|
842 |
|
|
843 |
(*Prove base case (subgoal i) and simplify others. A typical base case
|
|
844 |
concerns Crypt K X \<notin> Key`shrK`bad and cannot be proved by rewriting
|
|
845 |
alone.*)
|
|
846 |
fun prove_simple_subgoals_tac i =
|
|
847 |
force_tac (claset(), simpset() addsimps [image_eq_UN]) i THEN
|
|
848 |
ALLGOALS Asm_simp_tac
|
|
849 |
|
|
850 |
(*Analysis of Fake cases. Also works for messages that forward unknown parts,
|
|
851 |
but this application is no longer necessary if analz_insert_eq is used.
|
|
852 |
Abstraction over i is ESSENTIAL: it delays the dereferencing of claset
|
|
853 |
DEPENDS UPON "X" REFERRING TO THE FRADULENT MESSAGE *)
|
|
854 |
|
|
855 |
(*Apply rules to break down assumptions of the form
|
|
856 |
Y \<in> parts(insert X H) and Y \<in> analz(insert X H)
|
|
857 |
*)
|
|
858 |
val Fake_insert_tac =
|
|
859 |
dresolve_tac [impOfSubs Fake_analz_insert,
|
|
860 |
impOfSubs Fake_parts_insert] THEN'
|
|
861 |
eresolve_tac [asm_rl, thm"synth.Inj"];
|
|
862 |
|
|
863 |
fun Fake_insert_simp_tac ss i =
|
|
864 |
REPEAT (Fake_insert_tac i) THEN asm_full_simp_tac ss i;
|
|
865 |
|
|
866 |
fun atomic_spy_analz_tac (cs,ss) = SELECT_GOAL
|
|
867 |
(Fake_insert_simp_tac ss 1
|
|
868 |
THEN
|
|
869 |
IF_UNSOLVED (Blast.depth_tac
|
|
870 |
(cs addIs [analz_insertI,
|
|
871 |
impOfSubs analz_subset_parts]) 4 1))
|
|
872 |
|
|
873 |
(*The explicit claset and simpset arguments help it work with Isar*)
|
|
874 |
fun gen_spy_analz_tac (cs,ss) i =
|
|
875 |
DETERM
|
|
876 |
(SELECT_GOAL
|
|
877 |
(EVERY
|
|
878 |
[ (*push in occurrences of X...*)
|
|
879 |
(REPEAT o CHANGED)
|
|
880 |
(res_inst_tac [("x1","X")] (insert_commute RS ssubst) 1),
|
|
881 |
(*...allowing further simplifications*)
|
|
882 |
simp_tac ss 1,
|
|
883 |
REPEAT (FIRSTGOAL (resolve_tac [allI,impI,notI,conjI,iffI])),
|
|
884 |
DEPTH_SOLVE (atomic_spy_analz_tac (cs,ss) 1)]) i)
|
|
885 |
|
|
886 |
fun spy_analz_tac i = gen_spy_analz_tac (claset(), simpset()) i
|
|
887 |
*}
|
14218
|
888 |
(*>*)
|
|
889 |
|
14199
|
890 |
|
|
891 |
(*By default only o_apply is built-in. But in the presence of eta-expansion
|
|
892 |
this means that some terms displayed as (f o g) will be rewritten, and others
|
|
893 |
will not!*)
|
|
894 |
declare o_def [simp]
|
|
895 |
|
|
896 |
|
|
897 |
lemma Crypt_notin_image_Key [simp]: "Crypt K X \<notin> Key ` A"
|
|
898 |
by auto
|
|
899 |
|
|
900 |
lemma Hash_notin_image_Key [simp] :"Hash X \<notin> Key ` A"
|
|
901 |
by auto
|
|
902 |
|
|
903 |
lemma synth_analz_mono: "G<=H ==> synth (analz(G)) <= synth (analz(H))"
|
|
904 |
by (simp add: synth_mono analz_mono)
|
|
905 |
|
|
906 |
lemma Fake_analz_eq [simp]:
|
|
907 |
"X \<in> synth(analz H) ==> synth (analz (insert X H)) = synth (analz H)"
|
|
908 |
apply (drule Fake_analz_insert[of _ _ "H"])
|
|
909 |
apply (simp add: synth_increasing[THEN Un_absorb2])
|
|
910 |
apply (drule synth_mono)
|
|
911 |
apply (simp add: synth_idem)
|
|
912 |
apply (blast intro: synth_analz_mono [THEN [2] rev_subsetD])
|
|
913 |
done
|
|
914 |
|
|
915 |
text{*Two generalizations of @{text analz_insert_eq}*}
|
|
916 |
lemma gen_analz_insert_eq [rule_format]:
|
|
917 |
"X \<in> analz H ==> ALL G. H \<subseteq> G --> analz (insert X G) = analz G";
|
|
918 |
by (blast intro: analz_cut analz_insertI analz_mono [THEN [2] rev_subsetD])
|
|
919 |
|
|
920 |
lemma synth_analz_insert_eq [rule_format]:
|
|
921 |
"X \<in> synth (analz H)
|
|
922 |
==> ALL G. H \<subseteq> G --> (Key K \<in> analz (insert X G)) = (Key K \<in> analz G)";
|
|
923 |
apply (erule synth.induct)
|
|
924 |
apply (simp_all add: gen_analz_insert_eq subset_trans [OF _ subset_insertI])
|
|
925 |
done
|
|
926 |
|
|
927 |
lemma Fake_parts_sing:
|
|
928 |
"X \<in> synth (analz H) ==> parts{X} \<subseteq> synth (analz H) \<union> parts H";
|
|
929 |
apply (rule subset_trans)
|
|
930 |
apply (erule_tac [2] Fake_parts_insert)
|
|
931 |
apply (simp add: parts_mono)
|
|
932 |
done
|
|
933 |
|
|
934 |
lemmas Fake_parts_sing_imp_Un = Fake_parts_sing [THEN [2] rev_subsetD]
|
|
935 |
|
|
936 |
method_setup spy_analz = {*
|
|
937 |
Method.ctxt_args (fn ctxt =>
|
|
938 |
Method.METHOD (fn facts =>
|
15032
|
939 |
gen_spy_analz_tac (local_clasimpset_of ctxt) 1))*}
|
14199
|
940 |
"for proving the Fake case when analz is involved"
|
|
941 |
|
|
942 |
method_setup atomic_spy_analz = {*
|
|
943 |
Method.ctxt_args (fn ctxt =>
|
|
944 |
Method.METHOD (fn facts =>
|
15032
|
945 |
atomic_spy_analz_tac (local_clasimpset_of ctxt) 1))*}
|
14199
|
946 |
"for debugging spy_analz"
|
|
947 |
|
|
948 |
method_setup Fake_insert_simp = {*
|
|
949 |
Method.ctxt_args (fn ctxt =>
|
|
950 |
Method.METHOD (fn facts =>
|
15032
|
951 |
Fake_insert_simp_tac (local_simpset_of ctxt) 1))*}
|
14199
|
952 |
"for debugging spy_analz"
|
|
953 |
|
|
954 |
|
|
955 |
end
|