diff options
Diffstat (limited to 'meta/recipes-support/re2c/re2c/CVE-2018-21232-1.patch')
-rw-r--r-- | meta/recipes-support/re2c/re2c/CVE-2018-21232-1.patch | 347 |
1 files changed, 347 insertions, 0 deletions
diff --git a/meta/recipes-support/re2c/re2c/CVE-2018-21232-1.patch b/meta/recipes-support/re2c/re2c/CVE-2018-21232-1.patch new file mode 100644 index 0000000000..b7dcaefad3 --- /dev/null +++ b/meta/recipes-support/re2c/re2c/CVE-2018-21232-1.patch | |||
@@ -0,0 +1,347 @@ | |||
1 | From fd634998f813340768c333cdad638498602856e5 Mon Sep 17 00:00:00 2001 | ||
2 | From: Ulya Trofimovich <skvadrik@gmail.com> | ||
3 | Date: Tue, 21 Apr 2020 21:28:32 +0100 | ||
4 | Subject: [PATCH] Rewrite recursion into iteration (Tarjan's SCC algorithm and | ||
5 | YYFILL states). | ||
6 | |||
7 | This is to avoid stack overflow on large RE (especially on instrumented | ||
8 | builds that have larger stack frames, like AddressSanitizer). | ||
9 | |||
10 | Stack overflow reported by Agostino Sarubbo. | ||
11 | Related to #219 "overflow-1.re test fails on system with small stack". | ||
12 | |||
13 | Upstram-Status: Backport: | ||
14 | https://github.com/skvadrik/re2c/commit/fd634998f813340768c333cdad638498602856e5 | ||
15 | |||
16 | CVE: CVE-2018-21232 | ||
17 | |||
18 | Signed-off-by: Davide Gardenal <davide.gardenal@huawei.com> | ||
19 | --- | ||
20 | diff --git a/src/dfa/fillpoints.cc b/src/dfa/fillpoints.cc | ||
21 | --- a/src/dfa/fillpoints.cc (revision e58939b34bb4c37cd990f82dc286f21cb405743e) | ||
22 | +++ b/src/dfa/fillpoints.cc (date 1646929180243) | ||
23 | @@ -5,151 +5,186 @@ | ||
24 | |||
25 | #include "src/dfa/dfa.h" | ||
26 | |||
27 | -namespace re2c | ||
28 | -{ | ||
29 | + | ||
30 | +/* | ||
31 | + * note [finding strongly connected components of DFA] | ||
32 | + * | ||
33 | + * A slight modification of Tarjan's algorithm. | ||
34 | + * | ||
35 | + * The algorithm traverses the DFA in depth-first order. It maintains a stack | ||
36 | + * of states that have already been visited but haven't been assigned to an SCC | ||
37 | + * yet. For each state the algorithm calculates 'lowlink': index of the highest | ||
38 | + * ancestor state reachable in one step from a descendant of this state. | ||
39 | + * Lowlink is used to determine when a set of states should be popped off stack | ||
40 | + * into a new SCC. | ||
41 | + * | ||
42 | + * We use lowlink to hold different kinds of information: | ||
43 | + * - values in range [0 .. stack size] mean that the state is on stack (a | ||
44 | + * link to a state with the smallest index reachable from this one) | ||
45 | + * - SCC_UND means that this state has not been visited yet | ||
46 | + * - SCC_INF means that this state has already been popped off stack | ||
47 | + * | ||
48 | + * We use stack size (rather than topological sort index) as a unique index of | ||
49 | + * the state on stack. This is safe because the indices of states on stack are | ||
50 | + * unique and less than the indices of states that have been popped off stack | ||
51 | + * (SCC_INF). | ||
52 | + */ | ||
53 | + | ||
54 | +namespace re2c { | ||
55 | + namespace { | ||
56 | |||
57 | -static const size_t SCC_INF = std::numeric_limits<size_t>::max(); | ||
58 | -static const size_t SCC_UND = SCC_INF - 1; | ||
59 | + static const size_t SCC_INF = std::numeric_limits<size_t>::max(); | ||
60 | + static const size_t SCC_UND = SCC_INF - 1; | ||
61 | |||
62 | -static bool loopback(size_t node, size_t narcs, const size_t *arcs) | ||
63 | -{ | ||
64 | - for (size_t i = 0; i < narcs; ++i) | ||
65 | - { | ||
66 | - if (arcs[i] == node) | ||
67 | - { | ||
68 | - return true; | ||
69 | - } | ||
70 | - } | ||
71 | - return false; | ||
72 | -} | ||
73 | + static bool loopback(size_t state, size_t narcs, const size_t *arcs) | ||
74 | + { | ||
75 | + for (size_t i = 0; i < narcs; ++i) { | ||
76 | + if (arcs[i] == state) return true; | ||
77 | + } | ||
78 | + return false; | ||
79 | + } | ||
80 | |||
81 | -/* | ||
82 | - * node [finding strongly connected components of DFA] | ||
83 | - * | ||
84 | - * A slight modification of Tarjan's algorithm. | ||
85 | - * | ||
86 | - * The algorithm walks graph in deep-first order. It maintains a stack | ||
87 | - * of nodes that have already been visited but haven't been assigned to | ||
88 | - * SCC yet. For each node the algorithm calculates 'lowlink': index of | ||
89 | - * the highest ancestor node reachable in one step from a descendant of | ||
90 | - * the node. Lowlink is used to determine when a set of nodes should be | ||
91 | - * popped off the stack into a new SCC. | ||
92 | - * | ||
93 | - * We use lowlink to hold different kinds of information: | ||
94 | - * - values in range [0 .. stack size] mean that this node is on stack | ||
95 | - * (link to a node with the smallest index reachable from this one) | ||
96 | - * - SCC_UND means that this node has not been visited yet | ||
97 | - * - SCC_INF means that this node has already been popped off stack | ||
98 | - * | ||
99 | - * We use stack size (rather than topological sort index) as unique index | ||
100 | - * of a node on stack. This is safe because indices of nodes on stack are | ||
101 | - * still unique and less than indices of nodes that have been popped off | ||
102 | - * stack (SCC_INF). | ||
103 | - * | ||
104 | - */ | ||
105 | -static void scc( | ||
106 | - const dfa_t &dfa, | ||
107 | - std::stack<size_t> &stack, | ||
108 | - std::vector<size_t> &lowlink, | ||
109 | - std::vector<bool> &trivial, | ||
110 | - size_t i) | ||
111 | -{ | ||
112 | - const size_t link = stack.size(); | ||
113 | - lowlink[i] = link; | ||
114 | - stack.push(i); | ||
115 | + struct StackItem { | ||
116 | + size_t state; // current state | ||
117 | + size_t symbol; // next arc to be visited in this state | ||
118 | + size_t link; // Tarjan's "lowlink" | ||
119 | + }; | ||
120 | + | ||
121 | +// Tarjan's algorithm | ||
122 | + static void scc(const dfa_t &dfa, std::vector<bool> &trivial, | ||
123 | + std::vector<StackItem> &stack_dfs) | ||
124 | + { | ||
125 | + std::vector<size_t> lowlink(dfa.states.size(), SCC_UND); | ||
126 | + std::stack<size_t> stack; | ||
127 | + | ||
128 | + StackItem x0 = {0, 0, 0}; | ||
129 | + stack_dfs.push_back(x0); | ||
130 | + | ||
131 | + while (!stack_dfs.empty()) { | ||
132 | + const size_t i = stack_dfs.back().state; | ||
133 | + size_t c = stack_dfs.back().symbol; | ||
134 | + size_t link = stack_dfs.back().link; | ||
135 | + stack_dfs.pop_back(); | ||
136 | + | ||
137 | + const size_t *arcs = dfa.states[i]->arcs; | ||
138 | + | ||
139 | + if (c == 0) { | ||
140 | + // DFS recursive enter | ||
141 | + //DASSERT(lowlink[i] == SCC_UND); | ||
142 | + link = lowlink[i] = stack.size(); | ||
143 | + stack.push(i); | ||
144 | + } | ||
145 | + else { | ||
146 | + // DFS recursive return (from one of successor states) | ||
147 | + const size_t j = arcs[c - 1]; | ||
148 | + //DASSERT(lowlink[j] != SCC_UND); | ||
149 | + lowlink[i] = std::min(lowlink[i], lowlink[j]); | ||
150 | + } | ||
151 | |||
152 | - const size_t *arcs = dfa.states[i]->arcs; | ||
153 | - for (size_t c = 0; c < dfa.nchars; ++c) | ||
154 | - { | ||
155 | - const size_t j = arcs[c]; | ||
156 | - if (j != dfa_t::NIL) | ||
157 | - { | ||
158 | - if (lowlink[j] == SCC_UND) | ||
159 | - { | ||
160 | - scc(dfa, stack, lowlink, trivial, j); | ||
161 | - } | ||
162 | - if (lowlink[j] < lowlink[i]) | ||
163 | - { | ||
164 | - lowlink[i] = lowlink[j]; | ||
165 | - } | ||
166 | - } | ||
167 | - } | ||
168 | + // find the next successor state that hasn't been visited yet | ||
169 | + for (; c < dfa.nchars; ++c) { | ||
170 | + const size_t j = arcs[c]; | ||
171 | + if (j != dfa_t::NIL) { | ||
172 | + if (lowlink[j] == SCC_UND) { | ||
173 | + break; | ||
174 | + } | ||
175 | + lowlink[i] = std::min(lowlink[i], lowlink[j]); | ||
176 | + } | ||
177 | + } | ||
178 | |||
179 | - if (lowlink[i] == link) | ||
180 | - { | ||
181 | - // SCC is non-trivial (has loops) iff it either: | ||
182 | - // - consists of multiple nodes (they all must be interconnected) | ||
183 | - // - consists of single node which loops back to itself | ||
184 | - trivial[i] = i == stack.top() | ||
185 | - && !loopback(i, dfa.nchars, arcs); | ||
186 | + if (c < dfa.nchars) { | ||
187 | + // recurse into the next successor state | ||
188 | + StackItem x1 = {i, c + 1, link}; | ||
189 | + stack_dfs.push_back(x1); | ||
190 | + StackItem x2 = {arcs[c], 0, SCC_UND}; | ||
191 | + stack_dfs.push_back(x2); | ||
192 | + } | ||
193 | + else if (lowlink[i] == link) { | ||
194 | + // all successors have been visited | ||
195 | + // SCC is non-trivial (has loops) if either: | ||
196 | + // - it contains multiple interconnected states | ||
197 | + // - it contains a single self-looping state | ||
198 | + trivial[i] = i == stack.top() && !loopback(i, dfa.nchars, arcs); | ||
199 | |||
200 | - size_t j; | ||
201 | - do | ||
202 | - { | ||
203 | - j = stack.top(); | ||
204 | - stack.pop(); | ||
205 | - lowlink[j] = SCC_INF; | ||
206 | - } | ||
207 | - while (j != i); | ||
208 | - } | ||
209 | -} | ||
210 | + for (;;) { | ||
211 | + const size_t j = stack.top(); | ||
212 | + stack.pop(); | ||
213 | + lowlink[j] = SCC_INF; | ||
214 | + if (i == j) break; | ||
215 | + } | ||
216 | + } | ||
217 | + } | ||
218 | + } | ||
219 | |||
220 | -static void calc_fill( | ||
221 | - const dfa_t &dfa, | ||
222 | - const std::vector<bool> &trivial, | ||
223 | - std::vector<size_t> &fill, | ||
224 | - size_t i) | ||
225 | -{ | ||
226 | - if (fill[i] == SCC_UND) | ||
227 | - { | ||
228 | - fill[i] = 0; | ||
229 | - const size_t *arcs = dfa.states[i]->arcs; | ||
230 | - for (size_t c = 0; c < dfa.nchars; ++c) | ||
231 | - { | ||
232 | - const size_t j = arcs[c]; | ||
233 | - if (j != dfa_t::NIL) | ||
234 | - { | ||
235 | - calc_fill(dfa, trivial, fill, j); | ||
236 | - size_t max = 1; | ||
237 | - if (trivial[j]) | ||
238 | - { | ||
239 | - max += fill[j]; | ||
240 | - } | ||
241 | - if (max > fill[i]) | ||
242 | - { | ||
243 | - fill[i] = max; | ||
244 | - } | ||
245 | - } | ||
246 | - } | ||
247 | - } | ||
248 | -} | ||
249 | - | ||
250 | -void fillpoints(const dfa_t &dfa, std::vector<size_t> &fill) | ||
251 | -{ | ||
252 | - const size_t size = dfa.states.size(); | ||
253 | - | ||
254 | - // find DFA states that belong to non-trivial SCC | ||
255 | - std::stack<size_t> stack; | ||
256 | - std::vector<size_t> lowlink(size, SCC_UND); | ||
257 | - std::vector<bool> trivial(size, false); | ||
258 | - scc(dfa, stack, lowlink, trivial, 0); | ||
259 | - | ||
260 | - // for each DFA state, calculate YYFILL argument: | ||
261 | - // maximal path length to the next YYFILL state | ||
262 | - fill.resize(size, SCC_UND); | ||
263 | - calc_fill(dfa, trivial, fill, 0); | ||
264 | + static void calc_fill(const dfa_t &dfa, const std::vector<bool> &trivial, | ||
265 | + std::vector<StackItem> &stack_dfs, std::vector<size_t> &fill) | ||
266 | + { | ||
267 | + const size_t nstates = dfa.states.size(); | ||
268 | + fill.resize(nstates, SCC_UND); | ||
269 | + | ||
270 | + StackItem x0 = {0, 0, SCC_INF}; | ||
271 | + stack_dfs.push_back(x0); | ||
272 | + | ||
273 | + while (!stack_dfs.empty()) { | ||
274 | + const size_t i = stack_dfs.back().state; | ||
275 | + size_t c = stack_dfs.back().symbol; | ||
276 | + stack_dfs.pop_back(); | ||
277 | + | ||
278 | + const size_t *arcs = dfa.states[i]->arcs; | ||
279 | + | ||
280 | + if (c == 0) { | ||
281 | + // DFS recursive enter | ||
282 | + if (fill[i] != SCC_UND) continue; | ||
283 | + fill[i] = 0; | ||
284 | + } | ||
285 | + else { | ||
286 | + // DFS recursive return (from one of successor states) | ||
287 | + const size_t j = arcs[c - 1]; | ||
288 | + //DASSERT(fill[i] != SCC_UND && fill[j] != SCC_UND); | ||
289 | + fill[i] = std::max(fill[i], 1 + (trivial[j] ? fill[j] : 0)); | ||
290 | + } | ||
291 | + | ||
292 | + // find the next successor state that hasn't been visited yet | ||
293 | + for (; c < dfa.nchars; ++c) { | ||
294 | + const size_t j = arcs[c]; | ||
295 | + if (j != dfa_t::NIL) break; | ||
296 | + } | ||
297 | + | ||
298 | + if (c < dfa.nchars) { | ||
299 | + // recurse into the next successor state | ||
300 | + StackItem x1 = {i, c + 1, SCC_INF}; | ||
301 | + stack_dfs.push_back(x1); | ||
302 | + StackItem x2 = {arcs[c], 0, SCC_INF}; | ||
303 | + stack_dfs.push_back(x2); | ||
304 | + } | ||
305 | + } | ||
306 | |||
307 | - // The following states must trigger YYFILL: | ||
308 | - // - inital state | ||
309 | - // - all states in non-trivial SCCs | ||
310 | - // for other states, reset YYFILL argument to zero | ||
311 | - for (size_t i = 1; i < size; ++i) | ||
312 | - { | ||
313 | - if (trivial[i]) | ||
314 | - { | ||
315 | - fill[i] = 0; | ||
316 | - } | ||
317 | - } | ||
318 | -} | ||
319 | + // The following states must trigger YYFILL: | ||
320 | + // - inital state | ||
321 | + // - all states in non-trivial SCCs | ||
322 | + // for other states, reset YYFILL argument to zero | ||
323 | + for (size_t i = 1; i < nstates; ++i) { | ||
324 | + if (trivial[i]) { | ||
325 | + fill[i] = 0; | ||
326 | + } | ||
327 | + } | ||
328 | + } | ||
329 | |||
330 | + } // anonymous namespace | ||
331 | + | ||
332 | + void fillpoints(const dfa_t &dfa, std::vector<size_t> &fill) | ||
333 | + { | ||
334 | + const size_t nstates = dfa.states.size(); | ||
335 | + std::vector<bool> trivial(nstates, false); | ||
336 | + std::vector<StackItem> stack_dfs; | ||
337 | + stack_dfs.reserve(nstates); | ||
338 | + | ||
339 | + // find DFA states that belong to non-trivial SCC | ||
340 | + scc(dfa, trivial, stack_dfs); | ||
341 | + | ||
342 | + // for each DFA state, calculate YYFILL argument: | ||
343 | + // maximal path length to the next YYFILL state | ||
344 | + calc_fill(dfa, trivial, stack_dfs, fill); | ||
345 | + } | ||
346 | + | ||
347 | } // namespace re2c | ||