xref: /illumos-gate/usr/src/lib/libdemangle/common/rust-v0puny.c (revision 6f1fa39e3cf1b335f342bbca41590e9d76ab29b7)
1 /*
2  * This file and its contents are supplied under the terms of the
3  * Common Development and Distribution License ("CDDL"), version 1.0.
4  * You may only use this file in accordance with the terms of version
5  * 1.0 of the CDDL.
6  *
7  * A full copy of the text of the CDDL should have accompanied this
8  * source.  A copy of the CDDL is also available via the Internet at
9  * http://www.illumos.org/license/CDDL.
10  */
11 
12 /*
13  * Copyright 2019 Joyent, Inc.
14  * Copyright 2021 Jason King
15  */
16 
17 #include <inttypes.h>
18 #include <libcustr.h>
19 #include <limits.h>
20 #include <string.h>
21 #include <sys/byteorder.h>
22 #include "rust.h"
23 #include "strview.h"
24 
25 /*
26  * The rust v0 encoding (rust RFC 2603) uses a slightly modified
27  * version of punycode to encode characters that are not ASCII.
28  * The big difference is that '_' is used to separate the ASCII codepoints
29  * from the non-ASCII code points instead of '-'.
30  *
31  * The decoding is taken almost directly from (IETF) RFC 3492
32  */
33 
34 #define	BASE		36
35 #define	TMIN		1
36 #define	TMAX		26
37 #define	SKEW		38
38 #define	DAMP		700
39 #define	INITIAL_BIAS	72
40 #define	INITIAL_N	0x80
41 #define	DELIMITER	'_'
42 
43 static inline uint32_t char_val(char);
44 
45 static size_t
46 rustv0_puny_adapt(size_t delta, size_t npoints, boolean_t first)
47 {
48 	size_t k = 0;
49 
50 	delta = first ? delta / DAMP : delta / 2;
51 	delta += delta / npoints;
52 	while (delta > ((BASE - TMIN) * TMAX) / 2) {
53 		delta /= (BASE - TMIN);
54 		k += BASE;
55 	}
56 
57 	return (k + (((BASE - TMIN + 1) * delta) / (delta + SKEW)));
58 }
59 
60 boolean_t
61 rustv0_puny_decode(rust_state_t *restrict st, strview_t *restrict src,
62     boolean_t repl_underscore)
63 {
64 	uint32_t *buf;
65 	size_t bufalloc; /* in units of uint32_t */
66 	size_t buflen;
67 	size_t nbasic;
68 	size_t i, old_i, k, w;
69 	size_t n = INITIAL_N;
70 	size_t bias = INITIAL_BIAS;
71 	size_t delim_idx = 0;
72 	boolean_t ret = B_FALSE;
73 	char c;
74 
75 	DEMDEBUG("%s: str='%.*s'", __func__, SV_PRINT(src));
76 
77 	/*
78 	 * The decoded string should never contain more codepoints than
79 	 * the original string, so creating a temporary buffer large
80 	 * enought to hold sv_remaining(src) uint32_t's should be
81 	 * large enough.
82 	 *
83 	 * This also serves as a size check -- xcalloc will fail if the
84 	 * resulting size of the buf (sizeof (uint32_t) * bufalloc) >=
85 	 * SIZE_MAX. If xcalloc succeeds, we therefore know that that
86 	 * buflen cannot overflow.
87 	 */
88 	buflen = 0;
89 	bufalloc = sv_remaining(src) + 1;
90 	buf = xcalloc(st->rs_ops, bufalloc, sizeof (uint32_t));
91 	if (buf == NULL) {
92 		SET_ERROR(st);
93 		return (B_FALSE);
94 	}
95 
96 	/*
97 	 * Find the position of the last delimiter (if any).
98 	 * IETF RFC 3492 3.1 states that the delimiter is present if and only
99 	 * if there are a non-zero number of basic (ASCII) code points. Since
100 	 * the delimiter itself is a basic code point, the last one present
101 	 * in the original string is the actual delimiter between the basic
102 	 * and non-basic code points. Earlier occurences of the delimiter
103 	 * are treated as normal basic code points. For plain punycode, an
104 	 * all ASCII string encoded with punycode would terminate with a
105 	 * final delimiter, and a name with all non-basic code points would
106 	 * not have a delimiter at all. With the rust v0 encoding, punycode
107 	 * encoded identifiers have a 'u' prefix prior to the identifier
108 	 * length (['u'] <decimal-number> <bytes>), so we should never
109 	 * encounter an all ASCII name that's encoded with punycode (we error
110 	 * on this).  For an all non-basic codepoint identifier, no delimiter
111 	 * will be present, and we treat that the same as the delimiter being
112 	 * in the first position of the string, and consume it (if present)
113 	 * when we transition from copying the basic code points (which there
114 	 * will be none in this situation) to non-basic code points.
115 	 */
116 	for (i = 0; i < src->sv_rem; i++) {
117 		if (src->sv_first[i] == DELIMITER) {
118 			delim_idx = i;
119 		}
120 	}
121 	VERIFY3U(delim_idx, <, bufalloc);
122 
123 	if (delim_idx + 1 == sv_remaining(src)) {
124 		DEMDEBUG("%s: encountered an all-ASCII name encoded with "
125 		    "punycode", __func__);
126 		goto done;
127 	}
128 
129 	/* Copy all the basic characters up to the delimiter into buf */
130 	for (nbasic = 0; nbasic < delim_idx; nbasic++) {
131 		c = sv_consume_c(src);
132 
133 		/* The rust prefix check should guarantee this */
134 		VERIFY3U(c, <, 0x80);
135 
136 		/*
137 		 * Normal rust identifiers do not contain '-' in them.
138 		 * However ABI identifiers could contain a dash. Those
139 		 * are translated to _, and we need to replace accordingly
140 		 * when asked.
141 		 */
142 		if (repl_underscore && c == '_')
143 			c = '-';
144 
145 		buf[nbasic] = c;
146 		buflen++;
147 	}
148 	DEMDEBUG("%s: %" PRIu32 " ASCII codepoints copied", __func__, nbasic);
149 
150 	/*
151 	 * Consume delimiter between basic and non-basic code points if present.
152 	 * See above for explanation why it may not be present.
153 	 */
154 	(void) sv_consume_if_c(src, DELIMITER);
155 
156 	DEMDEBUG("%s: non-ASCII codepoints to decode: %.*s", __func__,
157 	    SV_PRINT(src));
158 
159 	for (i = 0; sv_remaining(src) > 0; i++) {
160 		VERIFY3U(i, <=, buflen);
161 
162 		/*
163 		 * Guarantee we have enough space to insert another codepoint.
164 		 * Our buffer sizing above should prevent this from ever
165 		 * tripping, but check this out of paranoia.
166 		 */
167 		VERIFY3U(buflen, <, bufalloc - 1);
168 
169 		/* decode the next codepoint */
170 		for (old_i = i, k = BASE, w = 1; ; k += BASE) {
171 			size_t t;
172 			uint32_t digit;
173 
174 			if (sv_remaining(src) == 0)
175 				goto done;
176 
177 			digit = char_val(sv_consume_c(src));
178 			if (digit >= BASE)
179 				goto done;
180 
181 			i = i + digit * w;
182 
183 			if (k <= bias)
184 				t = TMIN;
185 			else if (k >= bias + TMAX)
186 				t = TMAX;
187 			else
188 				t = k - bias;
189 
190 			if (digit < t)
191 				break;
192 
193 			w = w * (BASE - t);
194 		}
195 		buflen++;
196 
197 		bias = rustv0_puny_adapt(i - old_i, buflen,
198 		    (old_i == 0) ? B_TRUE : B_FALSE);
199 		n = n + i / buflen;
200 		i = i % buflen;
201 
202 		DEMDEBUG("%s: insert \\u%04" PRIx32 " at index %zu (len = %zu)",
203 		    __func__, n, i, buflen);
204 
205 		/*
206 		 * At the start of this while loop, we guaranteed
207 		 * buflen < bufalloc - 1. Therefore we know there is room
208 		 * to move over the contents of buf at i to make room
209 		 * for the codepoint. We also just guaranteed that i
210 		 * is in the range [0, buflen), so this should always be
211 		 * safe.
212 		 */
213 		(void) memmove(buf + i + 1, buf + i,
214 		    (buflen - i) * sizeof (uint32_t));
215 
216 #if _LP64
217 		/*
218 		 * This is always false for ILP32 and smatch will also complain,
219 		 * so we just omit it for ILP32.
220 		 */
221 		if (n > UINT32_MAX) {
222 			DEMDEBUG("%s: ERROR: utf8 value is out of range",
223 			    __func__);
224 			goto done;
225 		}
226 #endif
227 
228 		buf[i] = (uint32_t)n;
229 	}
230 
231 	DEMDEBUG("%s: inserted %zu non-basic code points", __func__,
232 	    buflen - nbasic);
233 
234 	for (i = 0; i < buflen; i++) {
235 		if (!rust_append_utf8_c(st, buf[i]))
236 			goto done;
237 	}
238 	ret = B_TRUE;
239 
240 done:
241 	xfree(st->rs_ops, buf, bufalloc * sizeof (uint32_t));
242 	return (ret);
243 }
244 
245 /*
246  * Convert [0-9][a-z] to a value [0..35]. Rust's punycode encoding always
247  * uses lowercase, so we treat uppercase (and any other characters) as
248  * invalid, and return BASE (36) to indicate a bad value.
249  */
250 static inline uint32_t
251 char_val(char c)
252 {
253 	uint32_t v = c;
254 
255 	if (ISLOWER(c)) {
256 		return (c - 'a');
257 	} else if (ISDIGIT(c)) {
258 		return (c - '0' + 26);
259 	} else {
260 		DEMDEBUG("%s: ERROR: invalid character 0x%02x encountered",
261 		    __func__, v);
262 		return (BASE);
263 	}
264 }
265