/* $NetBSD: regex.c,v 1.3.4.1 2024/02/29 12:35:03 martin Exp $ */ /* * Copyright (C) Internet Systems Consortium, Inc. ("ISC") * * SPDX-License-Identifier: MPL-2.0 * * This Source Code Form is subject to the terms of the Mozilla Public * License, v. 2.0. If a copy of the MPL was not distributed with this * file, you can obtain one at https://mozilla.org/MPL/2.0/. * * See the COPYRIGHT file distributed with this work for additional * information regarding copyright ownership. */ #include #include #include #include #include #if VALREGEX_REPORT_REASON #define FAIL(x) \ do { \ reason = (x); \ goto error; \ } while (0) #else /* if VALREGEX_REPORT_REASON */ #define FAIL(x) goto error #endif /* if VALREGEX_REPORT_REASON */ /* * Validate the regular expression 'C' locale. */ int isc_regex_validate(const char *c) { enum { none, parse_bracket, parse_bound, parse_ce, parse_ec, parse_cc } state = none; /* Well known character classes. */ const char *cc[] = { ":alnum:", ":digit:", ":punct:", ":alpha:", ":graph:", ":space:", ":blank:", ":lower:", ":upper:", ":cntrl:", ":print:", ":xdigit:" }; bool seen_comma = false; bool seen_high = false; bool seen_char = false; bool seen_ec = false; bool seen_ce = false; bool have_atom = false; int group = 0; int range = 0; int sub = 0; bool empty_ok = false; bool neg = false; bool was_multiple = false; unsigned int low = 0; unsigned int high = 0; const char *ccname = NULL; int range_start = 0; #if VALREGEX_REPORT_REASON const char *reason = ""; #endif /* if VALREGEX_REPORT_REASON */ if (c == NULL || *c == 0) { FAIL("empty string"); } while (c != NULL && *c != 0) { switch (state) { case none: switch (*c) { case '\\': /* make literal */ ++c; switch (*c) { case '1': case '2': case '3': case '4': case '5': case '6': case '7': case '8': case '9': if ((*c - '0') > sub) { FAIL("bad back reference"); } have_atom = true; was_multiple = false; break; case 0: FAIL("escaped end-of-string"); default: goto literal; } ++c; break; case '[': /* bracket start */ ++c; neg = false; was_multiple = false; seen_char = false; state = parse_bracket; break; case '{': /* bound start */ switch (c[1]) { case '0': case '1': case '2': case '3': case '4': case '5': case '6': case '7': case '8': case '9': if (!have_atom) { FAIL("no atom"); } if (was_multiple) { FAIL("was multiple"); } seen_comma = false; seen_high = false; low = high = 0; state = parse_bound; break; default: goto literal; } ++c; have_atom = true; was_multiple = true; break; case '}': goto literal; case '(': /* group start */ have_atom = false; was_multiple = false; empty_ok = true; ++group; ++sub; ++c; break; case ')': /* group end */ if (group && !have_atom && !empty_ok) { FAIL("empty alternative"); } have_atom = true; was_multiple = false; if (group != 0) { --group; } ++c; break; case '|': /* alternative separator */ if (!have_atom) { FAIL("no atom"); } have_atom = false; empty_ok = false; was_multiple = false; ++c; break; case '^': case '$': have_atom = true; was_multiple = true; ++c; break; case '+': case '*': case '?': if (was_multiple) { FAIL("was multiple"); } if (!have_atom) { FAIL("no atom"); } have_atom = true; was_multiple = true; ++c; break; case '.': default: literal: have_atom = true; was_multiple = false; ++c; break; } break; case parse_bound: switch (*c) { case '0': case '1': case '2': case '3': case '4': case '5': case '6': case '7': case '8': case '9': if (!seen_comma) { low = low * 10 + *c - '0'; if (low > 255) { FAIL("lower bound too big"); } } else { seen_high = true; high = high * 10 + *c - '0'; if (high > 255) { FAIL("upper bound too big"); } } ++c; break; case ',': if (seen_comma) { FAIL("multiple commas"); } seen_comma = true; ++c; break; default: case '{': FAIL("non digit/comma"); case '}': if (seen_high && low > high) { FAIL("bad parse bound"); } seen_comma = false; state = none; ++c; break; } break; case parse_bracket: switch (*c) { case '^': if (seen_char || neg) { goto inside; } neg = true; ++c; break; case '-': if (range == 2) { goto inside; } if (!seen_char) { goto inside; } if (range == 1) { FAIL("bad range"); } range = 2; ++c; break; case '[': ++c; switch (*c) { case '.': /* collating element */ if (range != 0) { --range; } ++c; state = parse_ce; seen_ce = false; break; case '=': /* equivalence class */ if (range == 2) { FAIL("equivalence class in " "range"); } ++c; state = parse_ec; seen_ec = false; break; case ':': /* character class */ if (range == 2) { FAIL("character class in " "range"); } ccname = c; ++c; state = parse_cc; break; } seen_char = true; break; case ']': if (!c[1] && !seen_char) { FAIL("unfinished brace"); } if (!seen_char) { goto inside; } ++c; range = 0; have_atom = true; state = none; break; default: inside: seen_char = true; if (range == 2 && (*c & 0xff) < range_start) { FAIL("out of order range"); } if (range != 0) { --range; } range_start = *c & 0xff; ++c; break; } break; case parse_ce: switch (*c) { case '.': ++c; switch (*c) { case ']': if (!seen_ce) { FAIL("empty ce"); } ++c; state = parse_bracket; break; default: if (seen_ce) { range_start = 256; } else { range_start = '.'; } seen_ce = true; break; } break; default: if (seen_ce) { range_start = 256; } else { range_start = *c; } seen_ce = true; ++c; break; } break; case parse_ec: switch (*c) { case '=': ++c; switch (*c) { case ']': if (!seen_ec) { FAIL("no ec"); } ++c; state = parse_bracket; break; default: seen_ec = true; break; } break; default: seen_ec = true; ++c; break; } break; case parse_cc: switch (*c) { case ':': ++c; switch (*c) { case ']': { unsigned int i; bool found = false; for (i = 0; i < sizeof(cc) / sizeof(*cc); i++) { unsigned int len; len = strlen(cc[i]); if (len != (unsigned int)(c - ccname)) { continue; } if (strncmp(cc[i], ccname, len)) { continue; } found = true; } if (!found) { FAIL("unknown cc"); } ++c; state = parse_bracket; break; } default: break; } break; default: ++c; break; } break; } } if (group != 0) { FAIL("group open"); } if (state != none) { FAIL("incomplete"); } if (!have_atom) { FAIL("no atom"); } return (sub); error: #if VALREGEX_REPORT_REASON fprintf(stderr, "%s\n", reason); #endif /* if VALREGEX_REPORT_REASON */ return (-1); }