| 1 | /* |
| 2 | * utf7.c - routines to handle UTF-7 (RFC 1642 / RFC 2152). |
| 3 | */ |
| 4 | |
| 5 | #ifndef ENUM_CHARSETS |
| 6 | |
| 7 | #include "charset.h" |
| 8 | #include "internal.h" |
| 9 | |
| 10 | /* |
| 11 | * This array is generated by a piece of Perl: |
| 12 | |
| 13 | perl -e 'for $i (0..32) { $a[$i] |= 2; } $a[32] |= 1;' \ |
| 14 | -e 'for $i ("a".."z","A".."Z","0".."9","'\''","(",' \ |
| 15 | -e ' ")",",","-",".","/",":","?") { $a[ord $i] |= 1; }' \ |
| 16 | -e 'for $i ("!","\"","#","\$","%","&","*",";","<","=",">","\@",' \ |
| 17 | -e ' "[","]","^","_","`","{","|","}") { $a[ord $i] |= 2; }' \ |
| 18 | -e 'for $i ("a".."z","A".."Z","0".."9","+","/") { $a[ord $i] |= 4; }' \ |
| 19 | -e 'for $i (0..127) { printf "%s%d,%s", $i%32?"":" ", $a[$i],' \ |
| 20 | -e ' ($i+1)%32?"":"\n"; }' |
| 21 | |
| 22 | */ |
| 23 | static const unsigned char utf7_ascii_properties[128] = { |
| 24 | 2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2, |
| 25 | 3,2,2,2,2,2,2,1,1,1,2,4,1,1,1,5,5,5,5,5,5,5,5,5,5,5,1,2,2,2,2,1, |
| 26 | 2,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,2,0,2,2,2, |
| 27 | 2,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,2,2,2,0,0, |
| 28 | }; |
| 29 | #define SET_D(c) ((c) >= 0 && (c) < 0x80 && (utf7_ascii_properties[(c)] & 1)) |
| 30 | #define SET_O(c) ((c) >= 0 && (c) < 0x80 && (utf7_ascii_properties[(c)] & 2)) |
| 31 | #define SET_B(c) ((c) >= 0 && (c) < 0x80 && (utf7_ascii_properties[(c)] & 4)) |
| 32 | |
| 33 | #define base64_value(c) ( (c) >= 'A' && (c) <= 'Z' ? (c) - 'A' : \ |
| 34 | (c) >= 'a' && (c) <= 'z' ? (c) - 'a' + 26 : \ |
| 35 | (c) >= '0' && (c) <= '9' ? (c) - '0' + 52 : \ |
| 36 | (c) == '+' ? 62 : 63 ) |
| 37 | |
| 38 | static const char *const base64_chars = |
| 39 | "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/"; |
| 40 | |
| 41 | static void read_utf7(charset_spec const *charset, long int input_chr, |
| 42 | charset_state *state, |
| 43 | void (*emit)(void *ctx, long int output), void *emitctx) |
| 44 | { |
| 45 | long int hw; |
| 46 | |
| 47 | UNUSEDARG(charset); |
| 48 | |
| 49 | /* |
| 50 | * state->s0 is used to handle the conversion of the UTF-7 |
| 51 | * transport format into a stream of halfwords. Its layout is: |
| 52 | * |
| 53 | * - In normal ASCII mode, it is zero. |
| 54 | * |
| 55 | * - Otherwise, it holds a leading 1 followed by all the bits |
| 56 | * so far accumulated in base64 digits. |
| 57 | * |
| 58 | * - Special case: when we have only just seen the initial `+' |
| 59 | * which enters base64 mode, it is set to 2 rather than 1 |
| 60 | * (this is an otherwise unused value since base64 always |
| 61 | * accumulates an even number of bits at a time), so that |
| 62 | * the special sequence `+-' can be made to encode `+' |
| 63 | * easily. |
| 64 | * |
| 65 | * state->s1 is used to handle the conversion of those |
| 66 | * halfwords into Unicode values. It contains a high surrogate |
| 67 | * value if we've just seen one, and 0 otherwise. |
| 68 | */ |
| 69 | |
| 70 | if (!state->s0) { |
| 71 | if (input_chr == '+') |
| 72 | state->s0 = 2; |
| 73 | else |
| 74 | emit(emitctx, input_chr); |
| 75 | return; |
| 76 | } else { |
| 77 | if (!SET_B(input_chr)) { |
| 78 | /* |
| 79 | * base64 mode ends here. Emit the character we have, |
| 80 | * unless it's a minus in which case we should swallow |
| 81 | * it. |
| 82 | */ |
| 83 | if (input_chr != '-') |
| 84 | emit(emitctx, input_chr); |
| 85 | else if (state->s0 == 2) |
| 86 | emit(emitctx, '+'); /* special case */ |
| 87 | state->s0 = 0; |
| 88 | return; |
| 89 | } |
| 90 | |
| 91 | /* |
| 92 | * Now we have a base64 character, so add it to our state, |
| 93 | * first correcting the special case value of s0. |
| 94 | */ |
| 95 | if (state->s0 == 2) |
| 96 | state->s0 = 1; |
| 97 | state->s0 = (state->s0 << 6) | base64_value(input_chr); |
| 98 | } |
| 99 | |
| 100 | /* |
| 101 | * If we don't have a whole halfword at this point, bale out. |
| 102 | */ |
| 103 | if (!(state->s0 & 0xFFFF0000)) |
| 104 | return; |
| 105 | |
| 106 | /* |
| 107 | * Otherwise, extract the halfword. There are three |
| 108 | * possibilities for where the top set bit might be. |
| 109 | */ |
| 110 | if (state->s0 & 0x00100000) { |
| 111 | hw = (state->s0 >> 4) & 0xFFFF; |
| 112 | state->s0 = (state->s0 & 0xF) | 0x10; |
| 113 | } else if (state->s0 & 0x00040000) { |
| 114 | hw = (state->s0 >> 2) & 0xFFFF; |
| 115 | state->s0 = (state->s0 & 3) | 4; |
| 116 | } else { |
| 117 | hw = state->s0 & 0xFFFF; |
| 118 | state->s0 = 1; |
| 119 | } |
| 120 | |
| 121 | /* |
| 122 | * Now what reaches this point should be a stream of halfwords |
| 123 | * in sensible numeric form. So now we process surrogates. |
| 124 | */ |
| 125 | if (state->s1) { |
| 126 | /* |
| 127 | * We have already seen a high surrogate, so we expect a |
| 128 | * low surrogate. Whinge if we didn't get it. |
| 129 | */ |
| 130 | if (hw < 0xDC00 || hw >= 0xE000) { |
| 131 | emit(emitctx, ERROR); |
| 132 | } else { |
| 133 | hw &= 0x3FF; |
| 134 | hw |= (state->s1 & 0x3FF) << 10; |
| 135 | emit(emitctx, hw + 0x10000); |
| 136 | } |
| 137 | state->s1 = 0; |
| 138 | } else { |
| 139 | /* |
| 140 | * Any low surrogate is an error. |
| 141 | */ |
| 142 | if (hw >= 0xDC00 && hw < 0xE000) { |
| 143 | emit(emitctx, ERROR); |
| 144 | return; |
| 145 | } |
| 146 | |
| 147 | /* |
| 148 | * Any high surrogate is simply stored until we see the |
| 149 | * next halfword. |
| 150 | */ |
| 151 | if (hw >= 0xD800 && hw < 0xDC00) { |
| 152 | state->s1 = hw; |
| 153 | return; |
| 154 | } |
| 155 | |
| 156 | /* |
| 157 | * Anything else we simply output. |
| 158 | */ |
| 159 | emit(emitctx, hw); |
| 160 | } |
| 161 | } |
| 162 | |
| 163 | /* |
| 164 | * For writing UTF-7, we supply two charset definitions, one of |
| 165 | * which will directly encode Set O characters and the other of |
| 166 | * which will cautiously base64 them. |
| 167 | */ |
| 168 | static int write_utf7(charset_spec const *charset, long int input_chr, |
| 169 | charset_state *state, |
| 170 | void (*emit)(void *ctx, long int output), |
| 171 | void *emitctx) |
| 172 | { |
| 173 | unsigned long hws[2]; |
| 174 | int nhws; |
| 175 | int i; |
| 176 | |
| 177 | /* |
| 178 | * For writing: state->s0 contains accumulated base64 data with |
| 179 | * a 1 in front, and state->s1 indicates how many bits of it we |
| 180 | * have. |
| 181 | */ |
| 182 | |
| 183 | if ((input_chr >= 0xD800 && input_chr < 0xE000) || |
| 184 | input_chr >= 0x110000) { |
| 185 | /* |
| 186 | * We can't output surrogates, or anything above 0x10FFFF. |
| 187 | */ |
| 188 | return FALSE; |
| 189 | } |
| 190 | |
| 191 | /* |
| 192 | * Look for characters which we output in ASCII mode. A special |
| 193 | * case here is +, which can be encoded as the empty base64 |
| 194 | * escape sequence `+-': if we're _already_ in ASCII mode we do |
| 195 | * that, but if we're in base64 mode at the point we see the + |
| 196 | * then we simply stay in base64 mode and output it as a |
| 197 | * halfword. (Switching back would cost three bytes, whereas |
| 198 | * staying in base64 costs only 2 2/3.) |
| 199 | */ |
| 200 | if (input_chr == -1 || SET_D(input_chr) || |
| 201 | (charset->charset == CS_UTF7 && SET_O(input_chr)) || |
| 202 | (!state->s0 && input_chr == '+')) { |
| 203 | if (state->s0) { |
| 204 | /* |
| 205 | * These characters are output in ASCII mode, so flush any |
| 206 | * lingering base64 data. |
| 207 | */ |
| 208 | state->s0 <<= 6 - state->s1; |
| 209 | emit(emitctx, base64_chars[state->s0 & 0x3F]); |
| 210 | /* |
| 211 | * I'm going to arbitrarily decide to always use the |
| 212 | * terminating minus sign. It's easier than figuring out |
| 213 | * whether to do so or not, and looks prettier besides. |
| 214 | */ |
| 215 | emit(emitctx, '-'); |
| 216 | state->s0 = state->s1 = 0; |
| 217 | } |
| 218 | |
| 219 | /* |
| 220 | * Now output the character. |
| 221 | */ |
| 222 | if (input_chr != -1) /* special case: just reset state */ |
| 223 | emit(emitctx, input_chr); |
| 224 | if (input_chr == '+') |
| 225 | emit(emitctx, '-'); /* +- encodes + */ |
| 226 | return TRUE; |
| 227 | } |
| 228 | |
| 229 | /* |
| 230 | * Now we know we have a character that needs to be output as |
| 231 | * either one base64-encoded halfword or two. So first figure |
| 232 | * out how many... |
| 233 | */ |
| 234 | if (input_chr < 0x10000) { |
| 235 | nhws = 1; |
| 236 | hws[0] = input_chr; |
| 237 | } else { |
| 238 | input_chr -= 0x10000; |
| 239 | if (input_chr >= 0x100000) { |
| 240 | /* Anything above 0x10FFFF is outside UTF-7 range. */ |
| 241 | return FALSE; |
| 242 | } |
| 243 | |
| 244 | nhws = 2; |
| 245 | hws[0] = 0xD800 | ((input_chr >> 10) & 0x3FF); |
| 246 | hws[1] = 0xDC00 | (input_chr & 0x3FF); |
| 247 | } |
| 248 | |
| 249 | /* |
| 250 | * ... switch into base64 mode if required ... |
| 251 | */ |
| 252 | if (!state->s0) { |
| 253 | emit(emitctx, '+'); |
| 254 | state->s0 = 1; |
| 255 | state->s1 = 0; |
| 256 | } |
| 257 | |
| 258 | /* |
| 259 | * ... and do the base64 output. |
| 260 | */ |
| 261 | for (i = 0; i < nhws; i++) { |
| 262 | state->s0 = (state->s0 << 16) | hws[i]; |
| 263 | state->s1 += 16; |
| 264 | |
| 265 | while (state->s1 >= 6) { |
| 266 | /* |
| 267 | * The top set bit must be in position 16, 18 or 20. |
| 268 | */ |
| 269 | unsigned long out, topbit; |
| 270 | |
| 271 | out = (state->s0 >> (state->s1 - 6)) & 0x3F; |
| 272 | state->s1 -= 6; |
| 273 | topbit = 1 << state->s1; |
| 274 | state->s0 = (state->s0 & (topbit-1)) | topbit; |
| 275 | |
| 276 | emit(emitctx, base64_chars[out]); |
| 277 | } |
| 278 | } |
| 279 | return TRUE; |
| 280 | } |
| 281 | |
| 282 | const charset_spec charset_CS_UTF7 = { |
| 283 | CS_UTF7, read_utf7, write_utf7, NULL |
| 284 | }; |
| 285 | |
| 286 | const charset_spec charset_CS_UTF7_CONSERVATIVE = { |
| 287 | CS_UTF7_CONSERVATIVE, read_utf7, write_utf7, NULL |
| 288 | }; |
| 289 | |
| 290 | #else /* ENUM_CHARSETS */ |
| 291 | |
| 292 | ENUM_CHARSET(CS_UTF7) |
| 293 | ENUM_CHARSET(CS_UTF7_CONSERVATIVE) |
| 294 | |
| 295 | #endif /* ENUM_CHARSETS */ |