/************************************************* * Perl-Compatible Regular Expressions * *************************************************/ /* This is a library of functions to support regular expressions whose syntax and semantics are as close as possible to those of the Perl 5 language. See the file Tech.Notes for some information on the internals. Written by: Philip Hazel Copyright (c) 1997-2004 University of Cambridge ----------------------------------------------------------------------------- Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: * Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer. * Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. * Neither the name of the University of Cambridge nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. ----------------------------------------------------------------------------- */ /* Define DEBUG to get debugging output on stdout. */ /* #define DEBUG */ /* Use a macro for debugging printing, 'cause that eliminates the use of #ifdef inline, and there are *still* stupid compilers about that don't like indented pre-processor statements. I suppose it's only been 10 years... */ #ifdef DEBUG #define DPRINTF(p) printf p #else #define DPRINTF(p) /*nothing*/ #endif /* Include the internals header, which itself includes "config.h", the Standard C headers, and the external pcre header. */ #include "internal.h" /* If Unicode Property support is wanted, include a private copy of the function that does it, and the table that translates names to numbers. */ #ifdef SUPPORT_UCP #include "ucp.c" #include "ucptypetable.c" #endif /* Maximum number of items on the nested bracket stacks at compile time. This applies to the nesting of all kinds of parentheses. It does not limit un-nested, non-capturing parentheses. This number can be made bigger if necessary - it is used to dimension one int and one unsigned char vector at compile time. */ #define BRASTACK_SIZE 200 /* Maximum number of ints of offset to save on the stack for recursive calls. If the offset vector is bigger, malloc is used. This should be a multiple of 3, because the offset vector is always a multiple of 3 long. */ #define REC_STACK_SAVE_MAX 30 /* The maximum remaining length of subject we are prepared to search for a req_byte match. */ #define REQ_BYTE_MAX 1000 /* Table of sizes for the fixed-length opcodes. It's defined in a macro so that the definition is next to the definition of the opcodes in internal.h. */ static const uschar OP_lengths[] = { OP_LENGTHS }; /* Min and max values for the common repeats; for the maxima, 0 => infinity */ static const char rep_min[] = { 0, 0, 1, 1, 0, 0 }; static const char rep_max[] = { 0, 0, 0, 0, 1, 1 }; /* Table for handling escaped characters in the range '0'-'z'. Positive returns are simple data values; negative values are for special things like \d and so on. Zero means further processing is needed (for things like \x), or the escape is invalid. */ #if !EBCDIC /* This is the "normal" table for ASCII systems */ static const short int escapes[] = { 0, 0, 0, 0, 0, 0, 0, 0, /* 0 - 7 */ 0, 0, ':', ';', '<', '=', '>', '?', /* 8 - ? */ '@', -ESC_A, -ESC_B, -ESC_C, -ESC_D, -ESC_E, 0, -ESC_G, /* @ - G */ 0, 0, 0, 0, 0, 0, 0, 0, /* H - O */ -ESC_P, -ESC_Q, 0, -ESC_S, 0, 0, 0, -ESC_W, /* P - W */ -ESC_X, 0, -ESC_Z, '[', '\\', ']', '^', '_', /* X - _ */ '`', 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0, /* ` - g */ 0, 0, 0, 0, 0, 0, ESC_n, 0, /* h - o */ -ESC_p, 0, ESC_r, -ESC_s, ESC_tee, 0, 0, -ESC_w, /* p - w */ 0, 0, -ESC_z /* x - z */ }; #else /* This is the "abnormal" table for EBCDIC systems */ static const short int escapes[] = { /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|', /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0, /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~', /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0, /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?', /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0, /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"', /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0, /* 88 */ 0, 0, 0, '{', 0, 0, 0, 0, /* 90 */ 0, 0, 0, 'l', 0, ESC_n, 0, -ESC_p, /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0, /* A0 */ 0, '~', -ESC_s, ESC_tee, 0, 0, -ESC_w, 0, /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0, /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0, /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-', /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G, /* C8 */ 0, 0, 0, 0, 0, 0, 0, 0, /* D0 */ '}', 0, 0, 0, 0, 0, 0, -ESC_P, /* D8 */-ESC_Q, 0, 0, 0, 0, 0, 0, 0, /* E0 */ '\\', 0, -ESC_S, 0, 0, 0, -ESC_W, -ESC_X, /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0, /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0, /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0 }; #endif /* Tables of names of POSIX character classes and their lengths. The list is terminated by a zero length entry. The first three must be alpha, upper, lower, as this is assumed for handling case independence. */ static const char *const posix_names[] = { "alpha", "lower", "upper", "alnum", "ascii", "blank", "cntrl", "digit", "graph", "print", "punct", "space", "word", "xdigit" }; static const uschar posix_name_lengths[] = { 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 }; /* Table of class bit maps for each POSIX class; up to three may be combined to form the class. The table for [:blank:] is dynamically modified to remove the vertical space characters. */ static const int posix_class_maps[] = { cbit_lower, cbit_upper, -1, /* alpha */ cbit_lower, -1, -1, /* lower */ cbit_upper, -1, -1, /* upper */ cbit_digit, cbit_lower, cbit_upper, /* alnum */ cbit_print, cbit_cntrl, -1, /* ascii */ cbit_space, -1, -1, /* blank - a GNU extension */ cbit_cntrl, -1, -1, /* cntrl */ cbit_digit, -1, -1, /* digit */ cbit_graph, -1, -1, /* graph */ cbit_print, -1, -1, /* print */ cbit_punct, -1, -1, /* punct */ cbit_space, -1, -1, /* space */ cbit_word, -1, -1, /* word - a Perl extension */ cbit_xdigit,-1, -1 /* xdigit */ }; /* Table to identify digits and hex digits. This is used when compiling patterns. Note that the tables in chartables are dependent on the locale, and may mark arbitrary characters as digits - but the PCRE compiling code expects to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have a private table here. It costs 256 bytes, but it is a lot faster than doing character value tests (at least in some simple cases I timed), and in some applications one wants PCRE to compile efficiently as well as match efficiently. For convenience, we use the same bit definitions as in chartables: 0x04 decimal digit 0x08 hexadecimal digit Then we can use ctype_digit and ctype_xdigit in the code. */ #if !EBCDIC /* This is the "normal" case, for ASCII systems */ static const unsigned char digitab[] = { 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */ 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */ 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */ 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */ 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */ #else /* This is the "abnormal" case, for EBCDIC systems */ static const unsigned char digitab[] = { 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- ¬ */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */ 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */ 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */ 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */ 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */ static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */ 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */ 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */ 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */ 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */ 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */ 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */ 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- ¬ */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */ 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */ 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */ 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */ 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */ 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */ 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */ 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */ 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */ 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */ 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */ 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */ 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */ 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */ 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */ 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */ 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */ 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */ 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */ #endif /* Definition to allow mutual recursion */ static PCRE_BOOL compile_regex(int, int, int *, uschar **, const uschar **, const char **, PCRE_BOOL, int, int *, int *, branch_chain *, compile_data *); /* Structure for building a chain of data that actually lives on the stack, for holding the values of the subject pointer at the start of each subpattern, so as to detect when an empty string has been matched by a subpattern - to break infinite loops. When NO_RECURSE is set, these blocks are on the heap, not on the stack. */ typedef struct eptrblock { struct eptrblock *epb_prev; const uschar *epb_saved_eptr; } eptrblock; /* Flag bits for the match() function */ #define match_condassert 0x01 /* Called to check a condition assertion */ #define match_isgroup 0x02 /* Set if start of bracketed group */ /* Non-error returns from the match() function. Error returns are externally defined PCRE_ERROR_xxx codes, which are all negative. */ #define MATCH_MATCH 1 #define MATCH_NOMATCH 0 /************************************************* * Global variables * *************************************************/ /* PCRE is thread-clean and doesn't use any global variables in the normal sense. However, it calls memory allocation and free functions via the four indirections below, and it can optionally do callouts. These values can be changed by the caller, but are shared between all threads. However, when compiling for Virtual Pascal, things are done differently (see pcre.in). */ #ifndef VPCOMPAT #ifdef __cplusplus extern "C" void *(*pcre_malloc)(size_t) = malloc; extern "C" void (*pcre_free)(void *) = free; extern "C" void *(*pcre_stack_malloc)(size_t) = malloc; extern "C" void (*pcre_stack_free)(void *) = free; extern "C" int (*pcre_callout)(pcre_callout_block *) = NULL; #else void *(*pcre_malloc)(size_t) = malloc; void (*pcre_free)(void *) = free; void *(*pcre_stack_malloc)(size_t) = malloc; void (*pcre_stack_free)(void *) = free; int (*pcre_callout)(pcre_callout_block *) = NULL; #endif #endif /************************************************* * Macros and tables for character handling * *************************************************/ /* When UTF-8 encoding is being used, a character is no longer just a single byte. The macros for character handling generate simple sequences when used in byte-mode, and more complicated ones for UTF-8 characters. */ #ifndef SUPPORT_UTF8 #define GETCHAR(c, eptr) c = *eptr; #define GETCHARINC(c, eptr) c = *eptr++; #define GETCHARINCTEST(c, eptr) c = *eptr++; #define GETCHARLEN(c, eptr, len) c = *eptr; #define BACKCHAR(eptr) #else /* SUPPORT_UTF8 */ /* Get the next UTF-8 character, not advancing the pointer. This is called when we know we are in UTF-8 mode. */ #define GETCHAR(c, eptr) \ c = *eptr; \ if ((c & 0xc0) == 0xc0) \ { \ int gcii; \ int gcaa = utf8_table4[c & 0x3f]; /* Number of additional bytes */ \ int gcss = 6*gcaa; \ c = (c & utf8_table3[gcaa]) << gcss; \ for (gcii = 1; gcii <= gcaa; gcii++) \ { \ gcss -= 6; \ c |= (eptr[gcii] & 0x3f) << gcss; \ } \ } /* Get the next UTF-8 character, advancing the pointer. This is called when we know we are in UTF-8 mode. */ #define GETCHARINC(c, eptr) \ c = *eptr++; \ if ((c & 0xc0) == 0xc0) \ { \ int gcaa = utf8_table4[c & 0x3f]; /* Number of additional bytes */ \ int gcss = 6*gcaa; \ c = (c & utf8_table3[gcaa]) << gcss; \ while (gcaa-- > 0) \ { \ gcss -= 6; \ c |= (*eptr++ & 0x3f) << gcss; \ } \ } /* Get the next character, testing for UTF-8 mode, and advancing the pointer */ #define GETCHARINCTEST(c, eptr) \ c = *eptr++; \ if (md->utf8 && (c & 0xc0) == 0xc0) \ { \ int gcaa = utf8_table4[c & 0x3f]; /* Number of additional bytes */ \ int gcss = 6*gcaa; \ c = (c & utf8_table3[gcaa]) << gcss; \ while (gcaa-- > 0) \ { \ gcss -= 6; \ c |= (*eptr++ & 0x3f) << gcss; \ } \ } /* Get the next UTF-8 character, not advancing the pointer, incrementing length if there are extra bytes. This is called when we know we are in UTF-8 mode. */ #define GETCHARLEN(c, eptr, len) \ c = *eptr; \ if ((c & 0xc0) == 0xc0) \ { \ int gcii; \ int gcaa = utf8_table4[c & 0x3f]; /* Number of additional bytes */ \ int gcss = 6*gcaa; \ c = (c & utf8_table3[gcaa]) << gcss; \ for (gcii = 1; gcii <= gcaa; gcii++) \ { \ gcss -= 6; \ c |= (eptr[gcii] & 0x3f) << gcss; \ } \ len += gcaa; \ } /* If the pointer is not at the start of a character, move it back until it is. Called only in UTF-8 mode. */ #define BACKCHAR(eptr) while((*eptr & 0xc0) == 0x80) eptr--; #endif /************************************************* * Default character tables * *************************************************/ /* A default set of character tables is included in the PCRE binary. Its source is built by the maketables auxiliary program, which uses the default C ctypes functions, and put in the file chartables.c. These tables are used by PCRE whenever the caller of pcre_compile() does not provide an alternate set of tables. */ #include "chartables.c" #ifdef SUPPORT_UTF8 /************************************************* * Tables for UTF-8 support * *************************************************/ /* These are the breakpoints for different numbers of bytes in a UTF-8 character. */ static const int utf8_table1[] = { 0x7f, 0x7ff, 0xffff, 0x1fffff, 0x3ffffff, 0x7fffffff}; /* These are the indicator bits and the mask for the data bits to set in the first byte of a character, indexed by the number of additional bytes. */ static const int utf8_table2[] = { 0, 0xc0, 0xe0, 0xf0, 0xf8, 0xfc}; static const int utf8_table3[] = { 0xff, 0x1f, 0x0f, 0x07, 0x03, 0x01}; /* Table of the number of extra characters, indexed by the first character masked with 0x3f. The highest number for a valid UTF-8 character is in fact 0x3d. */ static const uschar utf8_table4[] = { 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, 2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2, 3,3,3,3,3,3,3,3,4,4,4,4,5,5,5,5 }; /************************************************* * Convert character value to UTF-8 * *************************************************/ /* This function takes an integer value in the range 0 - 0x7fffffff and encodes it as a UTF-8 character in 0 to 6 bytes. Arguments: cvalue the character value buffer pointer to buffer for result - at least 6 bytes long Returns: number of characters placed in the buffer */ static int ord2utf8(int cvalue, uschar *buffer) { register int i, j; for (i = 0; i < sizeof(utf8_table1)/sizeof(int); i++) if (cvalue <= utf8_table1[i]) break; buffer += i; for (j = i; j > 0; j--) { *buffer-- = 0x80 | (cvalue & 0x3f); cvalue >>= 6; } *buffer = utf8_table2[i] | cvalue; return i + 1; } #endif /************************************************* * Print compiled regex * *************************************************/ /* The code for doing this is held in a separate file that is also included in pcretest.c. It defines a function called print_internals(). */ #ifdef DEBUG #include "printint.c" #endif /************************************************* * Return version string * *************************************************/ #define STRING(a) # a #define XSTRING(s) STRING(s) EXPORT const char * pcre_version(void) { return XSTRING(PCRE_MAJOR) "." XSTRING(PCRE_MINOR) " " XSTRING(PCRE_DATE); } /************************************************* * Flip bytes in an integer * *************************************************/ /* This function is called when the magic number in a regex doesn't match in order to flip its bytes to see if we are dealing with a pattern that was compiled on a host of different endianness. If so, this function is used to flip other byte values. Arguments: value the number to flip n the number of bytes to flip (assumed to be 2 or 4) Returns: the flipped value */ static long int byteflip(long int value, int n) { if (n == 2) return ((value & 0x00ff) << 8) | ((value & 0xff00) >> 8); return ((value & 0x000000ff) << 24) | ((value & 0x0000ff00) << 8) | ((value & 0x00ff0000) >> 8) | ((value & 0xff000000) >> 24); } /************************************************* * Test for a byte-flipped compiled regex * *************************************************/ /* This function is called from pce_exec() and also from pcre_fullinfo(). Its job is to test whether the regex is byte-flipped - that is, it was compiled on a system of opposite endianness. The function is called only when the native MAGIC_NUMBER test fails. If the regex is indeed flipped, we flip all the relevant values into a different data block, and return it. Arguments: re points to the regex study points to study data, or NULL internal_re points to a new regex block internal_study points to a new study block Returns: the new block if is is indeed a byte-flipped regex NULL if it is not */ static real_pcre * try_flipped(const real_pcre *re, real_pcre *internal_re, const pcre_study_data *study, pcre_study_data *internal_study) { if (byteflip(re->magic_number, sizeof(re->magic_number)) != MAGIC_NUMBER) return NULL; *internal_re = *re; /* To copy other fields */ internal_re->size = byteflip(re->size, sizeof(re->size)); internal_re->options = byteflip(re->options, sizeof(re->options)); internal_re->top_bracket = byteflip(re->top_bracket, sizeof(re->top_bracket)); internal_re->top_backref = byteflip(re->top_backref, sizeof(re->top_backref)); internal_re->first_byte = byteflip(re->first_byte, sizeof(re->first_byte)); internal_re->req_byte = byteflip(re->req_byte, sizeof(re->req_byte)); internal_re->name_table_offset = byteflip(re->name_table_offset, sizeof(re->name_table_offset)); internal_re->name_entry_size = byteflip(re->name_entry_size, sizeof(re->name_entry_size)); internal_re->name_count = byteflip(re->name_count, sizeof(re->name_count)); if (study != NULL) { *internal_study = *study; /* To copy other fields */ internal_study->size = byteflip(study->size, sizeof(study->size)); internal_study->options = byteflip(study->options, sizeof(study->options)); } return internal_re; } /************************************************* * (Obsolete) Return info about compiled pattern * *************************************************/ /* This is the original "info" function. It picks potentially useful data out of the private structure, but its interface was too rigid. It remains for backwards compatibility. The public options are passed back in an int - though the re->options field has been expanded to a long int, all the public options at the low end of it, and so even on 16-bit systems this will still be OK. Therefore, I haven't changed the API for pcre_info(). Arguments: argument_re points to compiled code optptr where to pass back the options first_byte where to pass back the first character, or -1 if multiline and all branches start ^, or -2 otherwise Returns: number of capturing subpatterns or negative values on error */ EXPORT int pcre_info(const pcre *argument_re, int *optptr, int *first_byte) { real_pcre internal_re; const real_pcre *re = (const real_pcre *)argument_re; if (re == NULL) return PCRE_ERROR_NULL; if (re->magic_number != MAGIC_NUMBER) { re = try_flipped(re, &internal_re, NULL, NULL); if (re == NULL) return PCRE_ERROR_BADMAGIC; } if (optptr != NULL) *optptr = (int)(re->options & PUBLIC_OPTIONS); if (first_byte != NULL) *first_byte = ((re->options & PCRE_FIRSTSET) != 0)? re->first_byte : ((re->options & PCRE_STARTLINE) != 0)? -1 : -2; return re->top_bracket; } /************************************************* * Return info about compiled pattern * *************************************************/ /* This is a newer "info" function which has an extensible interface so that additional items can be added compatibly. Arguments: argument_re points to compiled code extra_data points extra data, or NULL what what information is required where where to put the information Returns: 0 if data returned, negative on error */ EXPORT int pcre_fullinfo(const pcre *argument_re, const pcre_extra *extra_data, int what, void *where) { real_pcre internal_re; pcre_study_data internal_study; const real_pcre *re = (const real_pcre *)argument_re; const pcre_study_data *study = NULL; if (re == NULL || where == NULL) return PCRE_ERROR_NULL; if (extra_data != NULL && (extra_data->flags & PCRE_EXTRA_STUDY_DATA) != 0) study = (const pcre_study_data *)extra_data->study_data; if (re->magic_number != MAGIC_NUMBER) { re = try_flipped(re, &internal_re, study, &internal_study); if (re == NULL) return PCRE_ERROR_BADMAGIC; if (study != NULL) study = &internal_study; } switch (what) { case PCRE_INFO_OPTIONS: *((unsigned long int *)where) = re->options & PUBLIC_OPTIONS; break; case PCRE_INFO_SIZE: *((size_t *)where) = re->size; break; case PCRE_INFO_STUDYSIZE: *((size_t *)where) = (study == NULL)? 0 : study->size; break; case PCRE_INFO_CAPTURECOUNT: *((int *)where) = re->top_bracket; break; case PCRE_INFO_BACKREFMAX: *((int *)where) = re->top_backref; break; case PCRE_INFO_FIRSTBYTE: *((int *)where) = ((re->options & PCRE_FIRSTSET) != 0)? re->first_byte : ((re->options & PCRE_STARTLINE) != 0)? -1 : -2; break; /* Make sure we pass back the pointer to the bit vector in the external block, not the internal copy (with flipped integer fields). */ case PCRE_INFO_FIRSTTABLE: *((const uschar **)where) = (study != NULL && (study->options & PCRE_STUDY_MAPPED) != 0)? ((const pcre_study_data *)extra_data->study_data)->start_bits : NULL; break; case PCRE_INFO_LASTLITERAL: *((int *)where) = ((re->options & PCRE_REQCHSET) != 0)? re->req_byte : -1; break; case PCRE_INFO_NAMEENTRYSIZE: *((int *)where) = re->name_entry_size; break; case PCRE_INFO_NAMECOUNT: *((int *)where) = re->name_count; break; case PCRE_INFO_NAMETABLE: *((const uschar **)where) = (const uschar *)re + re->name_table_offset; break; case PCRE_INFO_DEFAULT_TABLES: *((const uschar **)where) = (const uschar *)pcre_default_tables; break; default: return PCRE_ERROR_BADOPTION; } return 0; } /************************************************* * Return info about what features are configured * *************************************************/ /* This is function which has an extensible interface so that additional items can be added compatibly. Arguments: what what information is required where where to put the information Returns: 0 if data returned, negative on error */ EXPORT int pcre_config(int what, void *where) { switch (what) { case PCRE_CONFIG_UTF8: #ifdef SUPPORT_UTF8 *((int *)where) = 1; #else *((int *)where) = 0; #endif break; case PCRE_CONFIG_UNICODE_PROPERTIES: #ifdef SUPPORT_UCP *((int *)where) = 1; #else *((int *)where) = 0; #endif break; case PCRE_CONFIG_NEWLINE: *((int *)where) = NEWLINE; break; case PCRE_CONFIG_LINK_SIZE: *((int *)where) = LINK_SIZE; break; case PCRE_CONFIG_POSIX_MALLOC_THRESHOLD: *((int *)where) = POSIX_MALLOC_THRESHOLD; break; case PCRE_CONFIG_MATCH_LIMIT: *((unsigned int *)where) = MATCH_LIMIT; break; case PCRE_CONFIG_STACKRECURSE: #ifdef NO_RECURSE *((int *)where) = 0; #else *((int *)where) = 1; #endif break; default: return PCRE_ERROR_BADOPTION; } return 0; } #ifdef DEBUG /************************************************* * Debugging function to print chars * *************************************************/ /* Print a sequence of chars in printable format, stopping at the end of the subject if the requested. Arguments: p points to characters length number to print is_subject TRUE if printing from within md->start_subject md pointer to matching data block, if is_subject is TRUE Returns: nothing */ static void pchars(const uschar *p, int length, PCRE_BOOL is_subject, match_data *md) { int c; if (is_subject && length > md->end_subject - p) length = md->end_subject - p; while (length-- > 0) if (isprint(c = *(p++))) printf("%c", c); else printf("\\x%02x", c); } #endif /************************************************* * Handle escapes * *************************************************/ /* This function is called when a \ has been encountered. It either returns a positive value for a simple escape such as \n, or a negative value which encodes one of the more complicated things such as \d. When UTF-8 is enabled, a positive value greater than 255 may be returned. On entry, ptr is pointing at the \. On exit, it is on the final character of the escape sequence. Arguments: ptrptr points to the pattern position pointer errorptr points to the pointer to the error message bracount number of previous extracting brackets options the options bits isclass TRUE if inside a character class Returns: zero or positive => a data character negative => a special escape sequence on error, errorptr is set */ static int check_escape(const uschar **ptrptr, const char **errorptr, int bracount, int options, PCRE_BOOL isclass) { const uschar *ptr = *ptrptr; int c, i; /* If backslash is at the end of the pattern, it's an error. */ c = *(++ptr); if (c == 0) *errorptr = ERR1; /* Non-alphamerics are literals. For digits or letters, do an initial lookup in a table. A non-zero result is something that can be returned immediately. Otherwise further processing may be required. */ #if !EBCDIC /* ASCII coding */ else if (c < '0' || c > 'z') {} /* Not alphameric */ else if ((i = escapes[c - '0']) != 0) c = i; #else /* EBCDIC coding */ else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphameric */ else if ((i = escapes[c - 0x48]) != 0) c = i; #endif /* Escapes that need further processing, or are illegal. */ else { const uschar *oldptr; switch (c) { /* A number of Perl escapes are not handled by PCRE. We give an explicit error. */ case 'l': case 'L': case 'N': case 'u': case 'U': *errorptr = ERR37; break; /* The handling of escape sequences consisting of a string of digits starting with one that is not zero is not straightforward. By experiment, the way Perl works seems to be as follows: Outside a character class, the digits are read as a decimal number. If the number is less than 10, or if there are that many previous extracting left brackets, then it is a back reference. Otherwise, up to three octal digits are read to form an escaped byte. Thus \123 is likely to be octal 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal value is greater than 377, the least significant 8 bits are taken. Inside a character class, \ followed by a digit is always an octal number. */ case '1': case '2': case '3': case '4': case '5': case '6': case '7': case '8': case '9': if (!isclass) { oldptr = ptr; c -= '0'; while ((digitab[ptr[1]] & ctype_digit) != 0) c = c * 10 + *(++ptr) - '0'; if (c < 10 || c <= bracount) { c = -(ESC_REF + c); break; } ptr = oldptr; /* Put the pointer back and fall through */ } /* Handle an octal number following \. If the first digit is 8 or 9, Perl generates a binary zero byte and treats the digit as a following literal. Thus we have to pull back the pointer by one. */ if ((c = *ptr) >= '8') { ptr--; c = 0; break; } /* \0 always starts an octal number, but we may drop through to here with a larger first octal digit. */ case '0': c -= '0'; while(i++ < 2 && ptr[1] >= '0' && ptr[1] <= '7') c = c * 8 + *(++ptr) - '0'; c &= 255; /* Take least significant 8 bits */ break; /* \x is complicated when UTF-8 is enabled. \x{ddd} is a character number which can be greater than 0xff, but only if the ddd are hex digits. */ case 'x': #ifdef SUPPORT_UTF8 if (ptr[1] == '{' && (options & PCRE_UTF8) != 0) { const uschar *pt = ptr + 2; register int count = 0; c = 0; while ((digitab[*pt] & ctype_xdigit) != 0) { int cc = *pt++; count++; #if !EBCDIC /* ASCII coding */ if (cc >= 'a') cc -= 32; /* Convert to upper case */ c = c * 16 + cc - ((cc < 'A')? '0' : ('A' - 10)); #else /* EBCDIC coding */ if (cc >= 'a' && cc <= 'z') cc += 64; /* Convert to upper case */ c = c * 16 + cc - ((cc >= '0')? '0' : ('A' - 10)); #endif } if (*pt == '}') { if (c < 0 || count > 8) *errorptr = ERR34; ptr = pt; break; } /* If the sequence of hex digits does not end with '}', then we don't recognize this construct; fall through to the normal \x handling. */ } #endif /* Read just a single hex char */ c = 0; while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0) { int cc; /* Some compilers don't like ++ */ cc = *(++ptr); /* in initializers */ #if !EBCDIC /* ASCII coding */ if (cc >= 'a') cc -= 32; /* Convert to upper case */ c = c * 16 + cc - ((cc < 'A')? '0' : ('A' - 10)); #else /* EBCDIC coding */ if (cc <= 'z') cc += 64; /* Convert to upper case */ c = c * 16 + cc - ((cc >= '0')? '0' : ('A' - 10)); #endif } break; /* Other special escapes not starting with a digit are straightforward */ case 'c': c = *(++ptr); if (c == 0) { *errorptr = ERR2; return 0; } /* A letter is upper-cased; then the 0x40 bit is flipped. This coding is ASCII-specific, but then the whole concept of \cx is ASCII-specific. (However, an EBCDIC equivalent has now been added.) */ #if !EBCDIC /* ASCII coding */ if (c >= 'a' && c <= 'z') c -= 32; c ^= 0x40; #else /* EBCDIC coding */ if (c >= 'a' && c <= 'z') c += 64; c ^= 0xC0; #endif break; /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any other alphameric following \ is an error if PCRE_EXTRA was set; otherwise, for Perl compatibility, it is a literal. This code looks a bit odd, but there used to be some cases other than the default, and there may be again in future, so I haven't "optimized" it. */ default: if ((options & PCRE_EXTRA) != 0) switch(c) { default: *errorptr = ERR3; break; } break; } } *ptrptr = ptr; return c; } #ifdef SUPPORT_UCP /************************************************* * Handle \P and \p * *************************************************/ /* This function is called after \P or \p has been encountered, provided that PCRE is compiled with support for Unicode properties. On entry, ptrptr is pointing at the P or p. On exit, it is pointing at the final character of the escape sequence. Argument: ptrptr points to the pattern position pointer negptr points to a boolean that is set TRUE for negation else FALSE errorptr points to the pointer to the error message Returns: value from ucp_type_table, or -1 for an invalid type */ static int get_ucp(const uschar **ptrptr, PCRE_BOOL *negptr, const char **errorptr) { int c, i, bot, top; const uschar *ptr = *ptrptr; char name[4]; c = *(++ptr); if (c == 0) goto ERROR_RETURN; *negptr = FALSE; /* \P or \p can be followed by a one- or two-character name in {}, optionally preceded by ^ for negation. */ if (c == '{') { if (ptr[1] == '^') { *negptr = TRUE; ptr++; } for (i = 0; i <= 2; i++) { c = *(++ptr); if (c == 0) goto ERROR_RETURN; if (c == '}') break; name[i] = c; } if (c !='}') /* Try to distinguish error cases */ { while (*(++ptr) != 0 && *ptr != '}'); if (*ptr == '}') goto UNKNOWN_RETURN; else goto ERROR_RETURN; } name[i] = 0; } /* Otherwise there is just one following character */ else { name[0] = c; name[1] = 0; } *ptrptr = ptr; /* Search for a recognized property name using binary chop */ bot = 0; top = sizeof(utt)/sizeof(ucp_type_table); while (bot < top) { i = (bot + top)/2; c = strcmp(name, utt[i].name); if (c == 0) return utt[i].value; if (c > 0) bot = i + 1; else top = i; } UNKNOWN_RETURN: *errorptr = ERR47; *ptrptr = ptr; return -1; ERROR_RETURN: *errorptr = ERR46; *ptrptr = ptr; return -1; } #endif /************************************************* * Check for counted repeat * *************************************************/ /* This function is called when a '{' is encountered in a place where it might start a quantifier. It looks ahead to see if it really is a quantifier or not. It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd} where the ddds are digits. Arguments: p pointer to the first char after '{' Returns: TRUE or FALSE */ static PCRE_BOOL is_counted_repeat(const uschar *p) { if ((digitab[*p++] & ctype_digit) == 0) return FALSE; while ((digitab[*p] & ctype_digit) != 0) p++; if (*p == '}') return TRUE; if (*p++ != ',') return FALSE; if (*p == '}') return TRUE; if ((digitab[*p++] & ctype_digit) == 0) return FALSE; while ((digitab[*p] & ctype_digit) != 0) p++; return (*p == '}'); } /************************************************* * Read repeat counts * *************************************************/ /* Read an item of the form {n,m} and return the values. This is called only after is_counted_repeat() has confirmed that a repeat-count quantifier exists, so the syntax is guaranteed to be correct, but we need to check the values. Arguments: p pointer to first char after '{' minp pointer to int for min maxp pointer to int for max returned as -1 if no max errorptr points to pointer to error message Returns: pointer to '}' on success; current ptr on error, with errorptr set */ static const uschar * read_repeat_counts(const uschar *p, int *minp, int *maxp, const char **errorptr) { int min = 0; int max = -1; while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - '0'; if (*p == '}') max = min; else { if (*(++p) != '}') { max = 0; while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - '0'; if (max < min) { *errorptr = ERR4; return p; } } } /* Do paranoid checks, then fill in the required variables, and pass back the pointer to the terminating '}'. */ if (min > 65535 || max > 65535) *errorptr = ERR5; else { *minp = min; *maxp = max; } return p; } /************************************************* * Find first significant op code * *************************************************/ /* This is called by several functions that scan a compiled expression looking for a fixed first character, or an anchoring op code etc. It skips over things that do not influence this. For some calls, a change of option is important. For some calls, it makes sense to skip negative forward and all backward assertions, and also the \b assertion; for others it does not. Arguments: code pointer to the start of the group options pointer to external options optbit the option bit whose changing is significant, or zero if none are skipassert TRUE if certain assertions are to be skipped Returns: pointer to the first significant opcode */ static const uschar* first_significant_code(const uschar *code, int *options, int optbit, PCRE_BOOL skipassert) { for (;;) { switch ((int)*code) { case OP_OPT: if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit)) *options = (int)code[1]; code += 2; break; case OP_ASSERT_NOT: case OP_ASSERTBACK: case OP_ASSERTBACK_NOT: if (!skipassert) return code; do code += GET(code, 1); while (*code == OP_ALT); code += OP_lengths[*code]; break; case OP_WORD_BOUNDARY: case OP_NOT_WORD_BOUNDARY: if (!skipassert) return code; /* Fall through */ case OP_CALLOUT: case OP_CREF: case OP_BRANUMBER: code += OP_lengths[*code]; break; default: return code; } } /* Control never reaches here */ } /************************************************* * Find the fixed length of a pattern * *************************************************/ /* Scan a pattern and compute the fixed length of subject that will match it, if the length is fixed. This is needed for dealing with backward assertions. In UTF8 mode, the result is in characters rather than bytes. Arguments: code points to the start of the pattern (the bracket) options the compiling options Returns: the fixed length, or -1 if there is no fixed length, or -2 if \C was encountered */ static int find_fixedlength(uschar *code, int options) { int length = -1; register int branchlength = 0; register uschar *cc = code + 1 + LINK_SIZE; /* Scan along the opcodes for this branch. If we get to the end of the branch, check the length against that of the other branches. */ for (;;) { int d; register int op = *cc; if (op >= OP_BRA) op = OP_BRA; switch (op) { case OP_BRA: case OP_ONCE: case OP_COND: d = find_fixedlength(cc, options); if (d < 0) return d; branchlength += d; do cc += GET(cc, 1); while (*cc == OP_ALT); cc += 1 + LINK_SIZE; break; /* Reached end of a branch; if it's a ket it is the end of a nested call. If it's ALT it is an alternation in a nested call. If it is END it's the end of the outer call. All can be handled by the same code. */ case OP_ALT: case OP_KET: case OP_KETRMAX: case OP_KETRMIN: case OP_END: if (length < 0) length = branchlength; else if (length != branchlength) return -1; if (*cc != OP_ALT) return length; cc += 1 + LINK_SIZE; branchlength = 0; break; /* Skip over assertive subpatterns */ case OP_ASSERT: case OP_ASSERT_NOT: case OP_ASSERTBACK: case OP_ASSERTBACK_NOT: do cc += GET(cc, 1); while (*cc == OP_ALT); /* Fall through */ /* Skip over things that don't match chars */ case OP_REVERSE: case OP_BRANUMBER: case OP_CREF: case OP_OPT: case OP_CALLOUT: case OP_SOD: case OP_SOM: case OP_EOD: case OP_EODN: case OP_CIRC: case OP_DOLL: case OP_NOT_WORD_BOUNDARY: case OP_WORD_BOUNDARY: cc += OP_lengths[*cc]; break; /* Handle literal characters */ case OP_CHAR: case OP_CHARNC: branchlength++; cc += 2; #ifdef SUPPORT_UTF8 if ((options & PCRE_UTF8) != 0) { while ((*cc & 0xc0) == 0x80) cc++; } #endif break; /* Handle exact repetitions. The count is already in characters, but we need to skip over a multibyte character in UTF8 mode. */ case OP_EXACT: branchlength += GET2(cc,1); cc += 4; #ifdef SUPPORT_UTF8 if ((options & PCRE_UTF8) != 0) { while((*cc & 0x80) == 0x80) cc++; } #endif break; case OP_TYPEEXACT: branchlength += GET2(cc,1); cc += 4; break; /* Handle single-char matchers */ case OP_PROP: case OP_NOTPROP: cc++; /* Fall through */ case OP_NOT_DIGIT: case OP_DIGIT: case OP_NOT_WHITESPACE: case OP_WHITESPACE: case OP_NOT_WORDCHAR: case OP_WORDCHAR: case OP_ANY: branchlength++; cc++; break; /* The single-byte matcher isn't allowed */ case OP_ANYBYTE: return -2; /* Check a class for variable quantification */ #ifdef SUPPORT_UTF8 case OP_XCLASS: cc += GET(cc, 1) - 33; /* Fall through */ #endif case OP_CLASS: case OP_NCLASS: cc += 33; switch (*cc) { case OP_CRSTAR: case OP_CRMINSTAR: case OP_CRQUERY: case OP_CRMINQUERY: return -1; case OP_CRRANGE: case OP_CRMINRANGE: if (GET2(cc,1) != GET2(cc,3)) return -1; branchlength += GET2(cc,1); cc += 5; break; default: branchlength++; } break; /* Anything else is variable length */ default: return -1; } } /* Control never gets here */ } /************************************************* * Scan compiled regex for numbered bracket * *************************************************/ /* This little function scans through a compiled pattern until it finds a capturing bracket with the given number. Arguments: code points to start of expression utf8 TRUE in UTF-8 mode number the required bracket number Returns: pointer to the opcode for the bracket, or NULL if not found */ static const uschar * find_bracket(const uschar *code, PCRE_BOOL utf8, int number) { #ifndef SUPPORT_UTF8 utf8 = utf8; /* Stop pedantic compilers complaining */ #endif for (;;) { register int c = *code; if (c == OP_END) return NULL; else if (c > OP_BRA) { int n = c - OP_BRA; if (n > EXTRACT_BASIC_MAX) n = GET2(code, 2+LINK_SIZE); if (n == number) return (uschar *)code; code += OP_lengths[OP_BRA]; } else { code += OP_lengths[c]; #ifdef SUPPORT_UTF8 /* In UTF-8 mode, opcodes that are followed by a character may be followed by a multi-byte character. The length in the table is a minimum, so we have to scan along to skip the extra bytes. All opcodes are less than 128, so we can use relatively efficient code. */ if (utf8) switch(c) { case OP_CHAR: case OP_CHARNC: case OP_EXACT: case OP_UPTO: case OP_MINUPTO: case OP_STAR: case OP_MINSTAR: case OP_PLUS: case OP_MINPLUS: case OP_QUERY: case OP_MINQUERY: while ((*code & 0xc0) == 0x80) code++; break; /* XCLASS is used for classes that cannot be represented just by a bit map. This includes negated single high-valued characters. The length in the table is zero; the actual length is stored in the compiled code. */ case OP_XCLASS: code += GET(code, 1) + 1; break; } #endif } } } /************************************************* * Scan compiled regex for recursion reference * *************************************************/ /* This little function scans through a compiled pattern until it finds an instance of OP_RECURSE. Arguments: code points to start of expression utf8 TRUE in UTF-8 mode Returns: pointer to the opcode for OP_RECURSE, or NULL if not found */ static const uschar * find_recurse(const uschar *code, PCRE_BOOL utf8) { #ifndef SUPPORT_UTF8 utf8 = utf8; /* Stop pedantic compilers complaining */ #endif for (;;) { register int c = *code; if (c == OP_END) return NULL; else if (c == OP_RECURSE) return code; else if (c > OP_BRA) { code += OP_lengths[OP_BRA]; } else { code += OP_lengths[c]; #ifdef SUPPORT_UTF8 /* In UTF-8 mode, opcodes that are followed by a character may be followed by a multi-byte character. The length in the table is a minimum, so we have to scan along to skip the extra bytes. All opcodes are less than 128, so we can use relatively efficient code. */ if (utf8) switch(c) { case OP_CHAR: case OP_CHARNC: case OP_EXACT: case OP_UPTO: case OP_MINUPTO: case OP_STAR: case OP_MINSTAR: case OP_PLUS: case OP_MINPLUS: case OP_QUERY: case OP_MINQUERY: while ((*code & 0xc0) == 0x80) code++; break; /* XCLASS is used for classes that cannot be represented just by a bit map. This includes negated single high-valued characters. The length in the table is zero; the actual length is stored in the compiled code. */ case OP_XCLASS: code += GET(code, 1) + 1; break; } #endif } } } /************************************************* * Scan compiled branch for non-emptiness * *************************************************/ /* This function scans through a branch of a compiled pattern to see whether it can match the empty string or not. It is called only from could_be_empty() below. Note that first_significant_code() skips over assertions. If we hit an unclosed bracket, we return "empty" - this means we've struck an inner bracket whose current branch will already have been scanned. Arguments: code points to start of search endcode points to where to stop utf8 TRUE if in UTF8 mode Returns: TRUE if what is matched could be empty */ static PCRE_BOOL could_be_empty_branch(const uschar *code, const uschar *endcode, PCRE_BOOL utf8) { register int c; for (code = first_significant_code(code + 1 + LINK_SIZE, NULL, 0, TRUE); code < endcode; code = first_significant_code(code + OP_lengths[c], NULL, 0, TRUE)) { const uschar *ccode; c = *code; if (c >= OP_BRA) { PCRE_BOOL empty_branch; if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */ /* Scan a closed bracket */ empty_branch = FALSE; do { if (!empty_branch && could_be_empty_branch(code, endcode, utf8)) empty_branch = TRUE; code += GET(code, 1); } while (*code == OP_ALT); if (!empty_branch) return FALSE; /* All branches are non-empty */ code += 1 + LINK_SIZE; c = *code; } else switch (c) { /* Check for quantifiers after a class */ #ifdef SUPPORT_UTF8 case OP_XCLASS: ccode = code + GET(code, 1); goto CHECK_CLASS_REPEAT; #endif case OP_CLASS: case OP_NCLASS: ccode = code + 33; #ifdef SUPPORT_UTF8 CHECK_CLASS_REPEAT: #endif switch (*ccode) { case OP_CRSTAR: /* These could be empty; continue */ case OP_CRMINSTAR: case OP_CRQUERY: case OP_CRMINQUERY: break; default: /* Non-repeat => class must match */ case OP_CRPLUS: /* These repeats aren't empty */ case OP_CRMINPLUS: return FALSE; case OP_CRRANGE: case OP_CRMINRANGE: if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */ break; } break; /* Opcodes that must match a character */ case OP_PROP: case OP_NOTPROP: case OP_EXTUNI: case OP_NOT_DIGIT: case OP_DIGIT: case OP_NOT_WHITESPACE: case OP_WHITESPACE: case OP_NOT_WORDCHAR: case OP_WORDCHAR: case OP_ANY: case OP_ANYBYTE: case OP_CHAR: case OP_CHARNC: case OP_NOT: case OP_PLUS: case OP_MINPLUS: case OP_EXACT: case OP_NOTPLUS: case OP_NOTMINPLUS: case OP_NOTEXACT: case OP_TYPEPLUS: case OP_TYPEMINPLUS: case OP_TYPEEXACT: return FALSE; /* End of branch */ case OP_KET: case OP_KETRMAX: case OP_KETRMIN: case OP_ALT: return TRUE; /* In UTF-8 mode, STAR, MINSTAR, QUERY, MINQUERY, UPTO, and MINUPTO may be followed by a multibyte character */ #ifdef SUPPORT_UTF8 case OP_STAR: case OP_MINSTAR: case OP_QUERY: case OP_MINQUERY: case OP_UPTO: case OP_MINUPTO: if (utf8) while ((code[2] & 0xc0) == 0x80) code++; break; #endif } } return TRUE; } /************************************************* * Scan compiled regex for non-emptiness * *************************************************/ /* This function is called to check for left recursive calls. We want to check the current branch of the current pattern to see if it could match the empty string. If it could, we must look outwards for branches at other levels, stopping when we pass beyond the bracket which is the subject of the recursion. Arguments: code points to start of the recursion endcode points to where to stop (current RECURSE item) bcptr points to the chain of current (unclosed) branch starts utf8 TRUE if in UTF-8 mode Returns: TRUE if what is matched could be empty */ static PCRE_BOOL could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr, PCRE_BOOL utf8) { while (bcptr != NULL && bcptr->current >= code) { if (!could_be_empty_branch(bcptr->current, endcode, utf8)) return FALSE; bcptr = bcptr->outer; } return TRUE; } /************************************************* * Check for POSIX class syntax * *************************************************/ /* This function is called when the sequence "[:" or "[." or "[=" is encountered in a character class. It checks whether this is followed by an optional ^ and then a sequence of letters, terminated by a matching ":]" or ".]" or "=]". Argument: ptr pointer to the initial [ endptr where to return the end pointer cd pointer to compile data Returns: TRUE or FALSE */ static PCRE_BOOL check_posix_syntax(const uschar *ptr, const uschar **endptr, compile_data *cd) { int terminator; /* Don't combine these lines; the Solaris cc */ terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */ if (*(++ptr) == '^') ptr++; while ((cd->ctypes[*ptr] & ctype_letter) != 0) ptr++; if (*ptr == terminator && ptr[1] == ']') { *endptr = ptr; return TRUE; } return FALSE; } /************************************************* * Check POSIX class name * *************************************************/ /* This function is called to check the name given in a POSIX-style class entry such as [:alnum:]. Arguments: ptr points to the first letter len the length of the name Returns: a value representing the name, or -1 if unknown */ static int check_posix_name(const uschar *ptr, int len) { register int yield = 0; while (posix_name_lengths[yield] != 0) { if (len == posix_name_lengths[yield] && strncmp((const char *)ptr, posix_names[yield], len) == 0) return yield; yield++; } return -1; } /************************************************* * Adjust OP_RECURSE items in repeated group * *************************************************/ /* OP_RECURSE items contain an offset from the start of the regex to the group that is referenced. This means that groups can be replicated for fixed repetition simply by copying (because the recursion is allowed to refer to earlier groups that are outside the current group). However, when a group is optional (i.e. the minimum quantifier is zero), OP_BRAZERO is inserted before it, after it has been compiled. This means that any OP_RECURSE items within it that refer to the group itself or any contained groups have to have their offsets adjusted. That is the job of this function. Before it is called, the partially compiled regex must be temporarily terminated with OP_END. Arguments: group points to the start of the group adjust the amount by which the group is to be moved utf8 TRUE in UTF-8 mode cd contains pointers to tables etc. Returns: nothing */ static void adjust_recurse(uschar *group, int adjust, PCRE_BOOL utf8, compile_data *cd) { uschar *ptr = group; while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL) { int offset = GET(ptr, 1); if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust); ptr += 1 + LINK_SIZE; } } /************************************************* * Insert an automatic callout point * *************************************************/ /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert callout points before each pattern item. Arguments: code current code pointer ptr current pattern pointer cd pointers to tables etc Returns: new code pointer */ static uschar * auto_callout(uschar *code, const uschar *ptr, compile_data *cd) { *code++ = OP_CALLOUT; *code++ = 255; PUT(code, 0, ptr - cd->start_pattern); /* Pattern offset */ PUT(code, LINK_SIZE, 0); /* Default length */ return code + 2*LINK_SIZE; } /************************************************* * Complete a callout item * *************************************************/ /* A callout item contains the length of the next item in the pattern, which we can't fill in till after we have reached the relevant point. This is used for both automatic and manual callouts. Arguments: previous_callout points to previous callout item ptr current pattern pointer cd pointers to tables etc Returns: nothing */ static void complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd) { int length = ptr - cd->start_pattern - GET(previous_callout, 2); PUT(previous_callout, 2 + LINK_SIZE, length); } #ifdef SUPPORT_UCP /************************************************* * Get othercase range * *************************************************/ /* This function is passed the start and end of a class range, in UTF-8 mode with UCP support. It searches up the characters, looking for internal ranges of characters in the "other" case. Each call returns the next one, updating the start address. Arguments: cptr points to starting character value; updated d end value ocptr where to put start of othercase range odptr where to put end of othercase range Yield: TRUE when range returned; FALSE when no more */ static PCRE_BOOL get_othercase_range(int *cptr, int d, int *ocptr, int *odptr) { int c, chartype, othercase, next; for (c = *cptr; c <= d; c++) { if (ucp_findchar(c, &chartype, &othercase) == ucp_L && othercase != 0) break; } if (c > d) return FALSE; *ocptr = othercase; next = othercase + 1; for (++c; c <= d; c++) { if (ucp_findchar(c, &chartype, &othercase) != ucp_L || othercase != next) break; next++; } *odptr = next - 1; *cptr = c; return TRUE; } #endif /* SUPPORT_UCP */ /************************************************* * Compile one branch * *************************************************/ /* Scan the pattern, compiling it into the code vector. If the options are changed during the branch, the pointer is used to change the external options bits. Arguments: optionsptr pointer to the option bits brackets points to number of extracting brackets used codeptr points to the pointer to the current code point ptrptr points to the current pattern pointer errorptr points to pointer to error message firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE) reqbyteptr set to the last literal character required, else < 0 bcptr points to current branch chain cd contains pointers to tables etc. Returns: TRUE on success FALSE, with *errorptr set on error */ static PCRE_BOOL compile_branch(int *optionsptr, int *brackets, uschar **codeptr, const uschar **ptrptr, const char **errorptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd) { int repeat_type, op_type; int repeat_min = 0, repeat_max = 0; /* To please picky compilers */ int bravalue = 0; int greedy_default, greedy_non_default; int firstbyte, reqbyte; int zeroreqbyte, zerofirstbyte; int req_caseopt, reqvary, tempreqvary; int condcount = 0; int options = *optionsptr; int after_manual_callout = 0; register int c; register uschar *code = *codeptr; uschar *tempcode; PCRE_BOOL inescq = FALSE; PCRE_BOOL groupsetfirstbyte = FALSE; const uschar *ptr = *ptrptr; const uschar *tempptr; uschar *previous = NULL; uschar *previous_callout = NULL; uschar classbits[32]; #ifdef SUPPORT_UTF8 PCRE_BOOL class_utf8; PCRE_BOOL utf8 = (options & PCRE_UTF8) != 0; uschar *class_utf8data; uschar utf8_char[6]; #else PCRE_BOOL utf8 = FALSE; #endif /* Set up the default and non-default settings for greediness */ greedy_default = ((options & PCRE_UNGREEDY) != 0); greedy_non_default = greedy_default ^ 1; /* Initialize no first byte, no required byte. REQ_UNSET means "no char matching encountered yet". It gets changed to REQ_NONE if we hit something that matches a non-fixed char first char; reqbyte just remains unset if we never find one. When we hit a repeat whose minimum is zero, we may have to adjust these values to take the zero repeat into account. This is implemented by setting them to zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual item types that can be repeated set these backoff variables appropriately. */ firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET; /* The variable req_caseopt contains either the REQ_CASELESS value or zero, according to the current setting of the caseless flag. REQ_CASELESS is a bit value > 255. It is added into the firstbyte or reqbyte variables to record the case status of the value. This is used only for ASCII characters. */ req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0; /* Switch on next character until the end of the branch */ for (;; ptr++) { PCRE_BOOL negate_class; PCRE_BOOL possessive_quantifier; PCRE_BOOL is_quantifier; int class_charcount; int class_lastchar; int newoptions; int recno; int skipbytes; int subreqbyte; int subfirstbyte; int mclength; uschar mcbuffer[8]; /* Next byte in the pattern */ c = *ptr; /* If in \Q...\E, check for the end; if not, we have a literal */ if (inescq && c != 0) { if (c == '\\' && ptr[1] == 'E') { inescq = FALSE; ptr++; continue; } else { if (previous_callout != NULL) { complete_callout(previous_callout, ptr, cd); previous_callout = NULL; } if ((options & PCRE_AUTO_CALLOUT) != 0) { previous_callout = code; code = auto_callout(code, ptr, cd); } goto NORMAL_CHAR; } } /* Fill in length of a previous callout, except when the next thing is a quantifier. */ is_quantifier = c == '*' || c == '+' || c == '?' || (c == '{' && is_counted_repeat(ptr+1)); if (!is_quantifier && previous_callout != NULL && after_manual_callout-- <= 0) { complete_callout(previous_callout, ptr, cd); previous_callout = NULL; } /* In extended mode, skip white space and comments */ if ((options & PCRE_EXTENDED) != 0) { if ((cd->ctypes[c] & ctype_space) != 0) continue; if (c == '#') { /* The space before the ; is to avoid a warning on a silly compiler on the Macintosh. */ while ((c = *(++ptr)) != 0 && c != NEWLINE) ; if (c != 0) continue; /* Else fall through to handle end of string */ } } /* No auto callout for quantifiers. */ if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier) { previous_callout = code; code = auto_callout(code, ptr, cd); } switch(c) { /* The branch terminates at end of string, |, or ). */ case 0: case '|': case ')': *firstbyteptr = firstbyte; *reqbyteptr = reqbyte; *codeptr = code; *ptrptr = ptr; return TRUE; /* Handle single-character metacharacters. In multiline mode, ^ disables the setting of any following char as a first character. */ case '^': if ((options & PCRE_MULTILINE) != 0) { if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE; } previous = NULL; *code++ = OP_CIRC; break; case '$': previous = NULL; *code++ = OP_DOLL; break; /* There can never be a first char if '.' is first, whatever happens about repeats. The value of reqbyte doesn't change either. */ case '.': if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE; zerofirstbyte = firstbyte; zeroreqbyte = reqbyte; previous = code; *code++ = OP_ANY; break; /* Character classes. If the included characters are all < 255 in value, we build a 32-byte bitmap of the permitted characters, except in the special case where there is only one such character. For negated classes, we build the map as usual, then invert it at the end. However, we use a different opcode so that data characters > 255 can be handled correctly. If the class contains characters outside the 0-255 range, a different opcode is compiled. It may optionally have a bit map for characters < 256, but those above are are explicitly listed afterwards. A flag byte tells whether the bitmap is present, and whether this is a negated class or not. */ case '[': previous = code; /* PCRE supports POSIX class stuff inside a class. Perl gives an error if they are encountered at the top level, so we'll do that too. */ if ((ptr[1] == ':' || ptr[1] == '.' || ptr[1] == '=') && check_posix_syntax(ptr, &tempptr, cd)) { *errorptr = (ptr[1] == ':')? ERR13 : ERR31; goto FAILED; } /* If the first character is '^', set the negation flag and skip it. */ if ((c = *(++ptr)) == '^') { negate_class = TRUE; c = *(++ptr); } else { negate_class = FALSE; } /* Keep a count of chars with values < 256 so that we can optimize the case of just a single character (as long as it's < 256). For higher valued UTF-8 characters, we don't yet do any optimization. */ class_charcount = 0; class_lastchar = -1; #ifdef SUPPORT_UTF8 class_utf8 = FALSE; /* No chars >= 256 */ class_utf8data = code + LINK_SIZE + 34; /* For UTF-8 items */ #endif /* Initialize the 32-char bit map to all zeros. We have to build the map in a temporary bit of store, in case the class contains only 1 character (< 256), because in that case the compiled code doesn't use the bit map. */ memset(classbits, 0, 32 * sizeof(uschar)); /* Process characters until ] is reached. By writing this as a "do" it means that an initial ] is taken as a data character. The first pass through the regex checked the overall syntax, so we don't need to be very strict here. At the start of the loop, c contains the first byte of the character. */ do { #ifdef SUPPORT_UTF8 if (utf8 && c > 127) { /* Braces are required because the */ GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */ } #endif /* Inside \Q...\E everything is literal except \E */ if (inescq) { if (c == '\\' && ptr[1] == 'E') { inescq = FALSE; ptr++; continue; } else goto LONE_SINGLE_CHARACTER; } /* Handle POSIX class names. Perl allows a negation extension of the form [:^name:]. A square bracket that doesn't match the syntax is treated as a literal. We also recognize the POSIX constructions [.ch.] and [=ch=] ("collating elements") and fault them, as Perl 5.6 and 5.8 do. */ if (c == '[' && (ptr[1] == ':' || ptr[1] == '.' || ptr[1] == '=') && check_posix_syntax(ptr, &tempptr, cd)) { PCRE_BOOL local_negate = FALSE; int posix_class, i; register const uschar *cbits = cd->cbits; if (ptr[1] != ':') { *errorptr = ERR31; goto FAILED; } ptr += 2; if (*ptr == '^') { local_negate = TRUE; ptr++; } posix_class = check_posix_name(ptr, tempptr - ptr); if (posix_class < 0) { *errorptr = ERR30; goto FAILED; } /* If matching is caseless, upper and lower are converted to alpha. This relies on the fact that the class table starts with alpha, lower, upper as the first 3 entries. */ if ((options & PCRE_CASELESS) != 0 && posix_class <= 2) posix_class = 0; /* Or into the map we are building up to 3 of the static class tables, or their negations. The [:blank:] class sets up the same chars as the [:space:] class (all white space). We remove the vertical white space chars afterwards. */ posix_class *= 3; for (i = 0; i < 3; i++) { PCRE_BOOL blankclass = strncmp((char *)ptr, "blank", 5) == 0; int taboffset = posix_class_maps[posix_class + i]; if (taboffset < 0) break; if (local_negate) { if (i == 0) for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+taboffset]; else for (c = 0; c < 32; c++) classbits[c] &= ~cbits[c+taboffset]; if (blankclass) classbits[1] |= 0x3c; } else { for (c = 0; c < 32; c++) classbits[c] |= cbits[c+taboffset]; if (blankclass) classbits[1] &= ~0x3c; } } ptr = tempptr + 1; class_charcount = 10; /* Set > 1; assumes more than 1 per class */ continue; /* End of POSIX syntax handling */ } /* Backslash may introduce a single character, or it may introduce one of the specials, which just set a flag. Escaped items are checked for validity in the pre-compiling pass. The sequence \b is a special case. Inside a class (and only there) it is treated as backspace. Elsewhere it marks a word boundary. Other escapes have preset maps ready to or into the one we are building. We assume they have more than one character in them, so set class_charcount bigger than one. */ if (c == '\\') { c = check_escape(&ptr, errorptr, *brackets, options, TRUE); if (-c == ESC_b) c = '\b'; /* \b is backslash in a class */ else if (-c == ESC_X) c = 'X'; /* \X is literal X in a class */ else if (-c == ESC_Q) /* Handle start of quoted string */ { if (ptr[1] == '\\' && ptr[2] == 'E') { ptr += 2; /* avoid empty string */ } else inescq = TRUE; continue; } if (c < 0) { register const uschar *cbits = cd->cbits; class_charcount += 2; /* Greater than 1 is what matters */ switch (-c) { case ESC_d: for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit]; continue; case ESC_D: for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit]; continue; case ESC_w: for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word]; continue; case ESC_W: for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word]; continue; case ESC_s: for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space]; classbits[1] &= ~0x08; /* Perl 5.004 onwards omits VT from \s */ continue; case ESC_S: for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space]; classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */ continue; #ifdef SUPPORT_UCP case ESC_p: case ESC_P: { PCRE_BOOL negated; int property = get_ucp(&ptr, &negated, errorptr); if (property < 0) goto FAILED; class_utf8 = TRUE; *class_utf8data++ = ((-c == ESC_p) != negated)? XCL_PROP : XCL_NOTPROP; *class_utf8data++ = property; class_charcount -= 2; /* Not a < 256 character */ } continue; #endif /* Unrecognized escapes are faulted if PCRE is running in its strict mode. By default, for compatibility with Perl, they are treated as literals. */ default: if ((options & PCRE_EXTRA) != 0) { *errorptr = ERR7; goto FAILED; } c = *ptr; /* The final character */ class_charcount -= 2; /* Undo the default count from above */ } } /* Fall through if we have a single character (c >= 0). This may be > 256 in UTF-8 mode. */ } /* End of backslash handling */ /* A single character may be followed by '-' to form a range. However, Perl does not permit ']' to be the end of the range. A '-' character here is treated as a literal. */ if (ptr[1] == '-' && ptr[2] != ']') { int d; ptr += 2; #ifdef SUPPORT_UTF8 if (utf8) { /* Braces are required because the */ GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */ } else #endif d = *ptr; /* Not UTF-8 mode */ /* The second part of a range can be a single-character escape, but not any of the other escapes. Perl 5.6 treats a hyphen as a literal in such circumstances. */ if (d == '\\') { const uschar *oldptr = ptr; d = check_escape(&ptr, errorptr, *brackets, options, TRUE); /* \b is backslash; \X is literal X; any other special means the '-' was literal */ if (d < 0) { if (d == -ESC_b) d = '\b'; else if (d == -ESC_X) d = 'X'; else { ptr = oldptr - 2; goto LONE_SINGLE_CHARACTER; /* A few lines below */ } } } /* The check that the two values are in the correct order happens in the pre-pass. Optimize one-character ranges */ if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */ /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless matching, we have to use an XCLASS with extra data items. Caseless matching for characters > 127 is available only if UCP support is available. */ #ifdef SUPPORT_UTF8 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127))) { class_utf8 = TRUE; /* With UCP support, we can find the other case equivalents of the relevant characters. There may be several ranges. Optimize how they fit with the basic range. */ #ifdef SUPPORT_UCP if ((options & PCRE_CASELESS) != 0) { int occ, ocd; int cc = c; int origd = d; while (get_othercase_range(&cc, origd, &occ, &ocd)) { if (occ >= c && ocd <= d) continue; /* Skip embedded ranges */ if (occ < c && ocd >= c - 1) /* Extend the basic range */ { /* if there is overlap, */ c = occ; /* noting that if occ < c */ continue; /* we can't have ocd > d */ } /* because a subrange is */ if (ocd > d && occ <= d + 1) /* always shorter than */ { /* the basic range. */ d = ocd; continue; } if (occ == ocd) { *class_utf8data++ = XCL_SINGLE; } else { *class_utf8data++ = XCL_RANGE; class_utf8data += ord2utf8(occ, class_utf8data); } class_utf8data += ord2utf8(ocd, class_utf8data); } } #endif /* SUPPORT_UCP */ /* Now record the original range, possibly modified for UCP caseless overlapping ranges. */ *class_utf8data++ = XCL_RANGE; class_utf8data += ord2utf8(c, class_utf8data); class_utf8data += ord2utf8(d, class_utf8data); /* With UCP support, we are done. Without UCP support, there is no caseless matching for UTF-8 characters > 127; we can use the bit map for the smaller ones. */ #ifdef SUPPORT_UCP continue; /* With next character in the class */ #else if ((options & PCRE_CASELESS) == 0 || c > 127) continue; /* Adjust upper limit and fall through to set up the map */ d = 127; #endif /* SUPPORT_UCP */ } #endif /* SUPPORT_UTF8 */ /* We use the bit map for all cases when not in UTF-8 mode; else ranges that lie entirely within 0-127 when there is UCP support; else for partial ranges without UCP support. */ for (; c <= d; c++) { classbits[c/8] |= (1 << (c&7)); if ((options & PCRE_CASELESS) != 0) { int uc = cd->fcc[c]; /* flip case */ classbits[uc/8] |= (1 << (uc&7)); } class_charcount++; /* in case a one-char range */ class_lastchar = c; } continue; /* Go get the next char in the class */ } /* Handle a lone single character - we can get here for a normal non-escape char, or after \ that introduces a single character or for an apparent range that isn't. */ LONE_SINGLE_CHARACTER: /* Handle a character that cannot go in the bit map */ #ifdef SUPPORT_UTF8 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127))) { class_utf8 = TRUE; *class_utf8data++ = XCL_SINGLE; class_utf8data += ord2utf8(c, class_utf8data); #ifdef SUPPORT_UCP if ((options & PCRE_CASELESS) != 0) { int chartype; int othercase; if (ucp_findchar(c, &chartype, &othercase) >= 0 && othercase > 0) { *class_utf8data++ = XCL_SINGLE; class_utf8data += ord2utf8(othercase, class_utf8data); } } #endif /* SUPPORT_UCP */ } else #endif /* SUPPORT_UTF8 */ /* Handle a single-byte character */ { classbits[c/8] |= (1 << (c&7)); if ((options & PCRE_CASELESS) != 0) { c = cd->fcc[c]; /* flip case */ classbits[c/8] |= (1 << (c&7)); } class_charcount++; class_lastchar = c; } } /* Loop until ']' reached; the check for end of string happens inside the loop. This "while" is the end of the "do" above. */ while ((c = *(++ptr)) != ']' || inescq); /* If class_charcount is 1, we saw precisely one character whose value is less than 256. In non-UTF-8 mode we can always optimize. In UTF-8 mode, we can optimize the negative case only if there were no characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR operate on single-bytes only. This is an historical hangover. Maybe one day we can tidy these opcodes to handle multi-byte characters. The optimization throws away the bit map. We turn the item into a 1-character OP_CHAR[NC] if it's positive, or OP_NOT if it's negative. Note that OP_NOT does not support multibyte characters. In the positive case, it can cause firstbyte to be set. Otherwise, there can be no first char if this item is first, whatever repeat count may follow. In the case of reqbyte, save the previous value for reinstating. */ #ifdef SUPPORT_UTF8 if (class_charcount == 1 && (!utf8 || (!class_utf8 && (!negate_class || class_lastchar < 128)))) #else if (class_charcount == 1) #endif { zeroreqbyte = reqbyte; /* The OP_NOT opcode works on one-byte characters only. */ if (negate_class) { if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE; zerofirstbyte = firstbyte; *code++ = OP_NOT; *code++ = class_lastchar; break; } /* For a single, positive character, get the value into mcbuffer, and then we can handle this with the normal one-character code. */ #ifdef SUPPORT_UTF8 if (utf8 && class_lastchar > 127) mclength = ord2utf8(class_lastchar, mcbuffer); else #endif { mcbuffer[0] = class_lastchar; mclength = 1; } goto ONE_CHAR; } /* End of 1-char optimization */ /* The general case - not the one-char optimization. If this is the first thing in the branch, there can be no first char setting, whatever the repeat count. Any reqbyte setting must remain unchanged after any kind of repeat. */ if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE; zerofirstbyte = firstbyte; zeroreqbyte = reqbyte; /* If there are characters with values > 255, we have to compile an extended class, with its own opcode. If there are no characters < 256, we can omit the bitmap. */ #ifdef SUPPORT_UTF8 if (class_utf8) { *class_utf8data++ = XCL_END; /* Marks the end of extra data */ *code++ = OP_XCLASS; code += LINK_SIZE; *code = negate_class? XCL_NOT : 0; /* If the map is required, install it, and move on to the end of the extra data */ if (class_charcount > 0) { *code++ |= XCL_MAP; memcpy(code, classbits, 32); code = class_utf8data; } /* If the map is not required, slide down the extra data. */ else { int len = class_utf8data - (code + 33); memmove(code + 1, code + 33, len); code += len + 1; } /* Now fill in the complete length of the item */ PUT(previous, 1, code - previous); break; /* End of class handling */ } #endif /* If there are no characters > 255, negate the 32-byte map if necessary, and copy it into the code vector. If this is the first thing in the branch, there can be no first char setting, whatever the repeat count. Any reqbyte setting must remain unchanged after any kind of repeat. */ if (negate_class) { *code++ = OP_NCLASS; for (c = 0; c < 32; c++) code[c] = ~classbits[c]; } else { *code++ = OP_CLASS; memcpy(code, classbits, 32); } code += 32; break; /* Various kinds of repeat; '{' is not necessarily a quantifier, but this has been tested above. */ case '{': if (!is_quantifier) goto NORMAL_CHAR; ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorptr); if (*errorptr != NULL) goto FAILED; goto REPEAT; case '*': repeat_min = 0; repeat_max = -1; goto REPEAT; case '+': repeat_min = 1; repeat_max = -1; goto REPEAT; case '?': repeat_min = 0; repeat_max = 1; REPEAT: if (previous == NULL) { *errorptr = ERR9; goto FAILED; } if (repeat_min == 0) { firstbyte = zerofirstbyte; /* Adjust for zero repeat */ reqbyte = zeroreqbyte; /* Ditto */ } /* Remember whether this is a variable length repeat */ reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY; op_type = 0; /* Default single-char op codes */ possessive_quantifier = FALSE; /* Default not possessive quantifier */ /* Save start of previous item, in case we have to move it up to make space for an inserted OP_ONCE for the additional '+' extension. */ tempcode = previous; /* If the next character is '+', we have a possessive quantifier. This implies greediness, whatever the setting of the PCRE_UNGREEDY option. If the next character is '?' this is a minimizing repeat, by default, but if PCRE_UNGREEDY is set, it works the other way round. We change the repeat type to the non-default. */ if (ptr[1] == '+') { repeat_type = 0; /* Force greedy */ possessive_quantifier = TRUE; ptr++; } else if (ptr[1] == '?') { repeat_type = greedy_non_default; ptr++; } else repeat_type = greedy_default; /* If previous was a recursion, we need to wrap it inside brackets so that it can be replicated if necessary. */ if (*previous == OP_RECURSE) { memmove(previous + 1 + LINK_SIZE, previous, 1 + LINK_SIZE); code += 1 + LINK_SIZE; *previous = OP_BRA; PUT(previous, 1, code - previous); *code = OP_KET; PUT(code, 1, code - previous); code += 1 + LINK_SIZE; } /* If previous was a character match, abolish the item and generate a repeat item instead. If a char item has a minumum of more than one, ensure that it is set in reqbyte - it might not be if a sequence such as x{3} is the first thing in a branch because the x will have gone into firstbyte instead. */ if (*previous == OP_CHAR || *previous == OP_CHARNC) { /* Deal with UTF-8 characters that take up more than one byte. It's easier to write this out separately than try to macrify it. Use c to hold the length of the character in bytes, plus 0x80 to flag that it's a length rather than a small character. */ #ifdef SUPPORT_UTF8 if (utf8 && (code[-1] & 0x80) != 0) { uschar *lastchar = code - 1; while((*lastchar & 0xc0) == 0x80) lastchar--; c = code - lastchar; /* Length of UTF-8 character */ memcpy(utf8_char, lastchar, c); /* Save the char */ c |= 0x80; /* Flag c as a length */ } else #endif /* Handle the case of a single byte - either with no UTF8 support, or with UTF-8 disabled, or for a UTF-8 character < 128. */ { c = code[-1]; if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt; } goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */ } /* If previous was a single negated character ([^a] or similar), we use one of the special opcodes, replacing it. The code is shared with single- character repeats by setting opt_type to add a suitable offset into repeat_type. OP_NOT is currently used only for single-byte chars. */ else if (*previous == OP_NOT) { op_type = OP_NOTSTAR - OP_STAR; /* Use "not" opcodes */ c = previous[1]; goto OUTPUT_SINGLE_REPEAT; } /* If previous was a character type match (\d or similar), abolish it and create a suitable repeat item. The code is shared with single-character repeats by setting op_type to add a suitable offset into repeat_type. Note the the Unicode property types will be present only when SUPPORT_UCP is defined, but we don't wrap the little bits of code here because it just makes it horribly messy. */ else if (*previous < OP_EODN) { uschar *oldcode; int prop_type; op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */ c = *previous; OUTPUT_SINGLE_REPEAT: prop_type = (*previous == OP_PROP || *previous == OP_NOTPROP)? previous[1] : -1; oldcode = code; code = previous; /* Usually overwrite previous item */ /* If the maximum is zero then the minimum must also be zero; Perl allows this case, so we do too - by simply omitting the item altogether. */ if (repeat_max == 0) goto END_REPEAT; /* All real repeats make it impossible to handle partial matching (maybe one day we will be able to remove this restriction). */ if (repeat_max != 1) cd->nopartial = TRUE; /* Combine the op_type with the repeat_type */ repeat_type += op_type; /* A minimum of zero is handled either as the special case * or ?, or as an UPTO, with the maximum given. */ if (repeat_min == 0) { if (repeat_max == -1) *code++ = OP_STAR + repeat_type; else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type; else { *code++ = OP_UPTO + repeat_type; PUT2INC(code, 0, repeat_max); } } /* A repeat minimum of 1 is optimized into some special cases. If the maximum is unlimited, we use OP_PLUS. Otherwise, the original item it left in place and, if the maximum is greater than 1, we use OP_UPTO with one less than the maximum. */ else if (repeat_min == 1) { if (repeat_max == -1) *code++ = OP_PLUS + repeat_type; else { code = oldcode; /* leave previous item in place */ if (repeat_max == 1) goto END_REPEAT; *code++ = OP_UPTO + repeat_type; PUT2INC(code, 0, repeat_max - 1); } } /* The case {n,n} is just an EXACT, while the general case {n,m} is handled as an EXACT followed by an UPTO. */ else { *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */ PUT2INC(code, 0, repeat_min); /* If the maximum is unlimited, insert an OP_STAR. Before doing so, we have to insert the character for the previous code. For a repeated Unicode property match, there is an extra byte that defines the required property. In UTF-8 mode, long characters have their length in c, with the 0x80 bit as a flag. */ if (repeat_max < 0) { #ifdef SUPPORT_UTF8 if (utf8 && c >= 128) { memcpy(code, utf8_char, c & 7); code += c & 7; } else #endif { *code++ = c; if (prop_type >= 0) *code++ = prop_type; } *code++ = OP_STAR + repeat_type; } /* Else insert an UPTO if the max is greater than the min, again preceded by the character, for the previously inserted code. */ else if (repeat_max != repeat_min) { #ifdef SUPPORT_UTF8 if (utf8 && c >= 128) { memcpy(code, utf8_char, c & 7); code += c & 7; } else #endif *code++ = c; if (prop_type >= 0) *code++ = prop_type; repeat_max -= repeat_min; *code++ = OP_UPTO + repeat_type; PUT2INC(code, 0, repeat_max); } } /* The character or character type itself comes last in all cases. */ #ifdef SUPPORT_UTF8 if (utf8 && c >= 128) { memcpy(code, utf8_char, c & 7); code += c & 7; } else #endif *code++ = c; /* For a repeated Unicode property match, there is an extra byte that defines the required property. */ #ifdef SUPPORT_UCP if (prop_type >= 0) *code++ = prop_type; #endif } /* If previous was a character class or a back reference, we put the repeat stuff after it, but just skip the item if the repeat was {0,0}. */ else if (*previous == OP_CLASS || *previous == OP_NCLASS || #ifdef SUPPORT_UTF8 *previous == OP_XCLASS || #endif *previous == OP_REF) { if (repeat_max == 0) { code = previous; goto END_REPEAT; } /* All real repeats make it impossible to handle partial matching (maybe one day we will be able to remove this restriction). */ if (repeat_max != 1) cd->nopartial = TRUE; if (repeat_min == 0 && repeat_max == -1) *code++ = OP_CRSTAR + repeat_type; else if (repeat_min == 1 && repeat_max == -1) *code++ = OP_CRPLUS + repeat_type; else if (repeat_min == 0 && repeat_max == 1) *code++ = OP_CRQUERY + repeat_type; else { *code++ = OP_CRRANGE + repeat_type; PUT2INC(code, 0, repeat_min); if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */ PUT2INC(code, 0, repeat_max); } } /* If previous was a bracket group, we may have to replicate it in certain cases. */ else if (*previous >= OP_BRA || *previous == OP_ONCE || *previous == OP_COND) { register int i; int ketoffset = 0; int len = code - previous; uschar *bralink = NULL; /* If the maximum repeat count is unlimited, find the end of the bracket by scanning through from the start, and compute the offset back to it from the current code pointer. There may be an OP_OPT setting following the final KET, so we can't find the end just by going back from the code pointer. */ if (repeat_max == -1) { register uschar *ket = previous; do ket += GET(ket, 1); while (*ket != OP_KET); ketoffset = code - ket; } /* The case of a zero minimum is special because of the need to stick OP_BRAZERO in front of it, and because the group appears once in the data, whereas in other cases it appears the minimum number of times. For this reason, it is simplest to treat this case separately, as otherwise the code gets far too messy. There are several special subcases when the minimum is zero. */ if (repeat_min == 0) { /* If the maximum is also zero, we just omit the group from the output altogether. */ if (repeat_max == 0) { code = previous; goto END_REPEAT; } /* If the maximum is 1 or unlimited, we just have to stick in the BRAZERO and do no more at this point. However, we do need to adjust any OP_RECURSE calls inside the group that refer to the group itself or any internal group, because the offset is from the start of the whole regex. Temporarily terminate the pattern while doing this. */ if (repeat_max <= 1) { *code = OP_END; adjust_recurse(previous, 1, utf8, cd); memmove(previous+1, previous, len); code++; *previous++ = OP_BRAZERO + repeat_type; } /* If the maximum is greater than 1 and limited, we have to replicate in a nested fashion, sticking OP_BRAZERO before each set of brackets. The first one has to be handled carefully because it's the original copy, which has to be moved up. The remainder can be handled by code that is common with the non-zero minimum case below. We have to adjust the value or repeat_max, since one less copy is required. Once again, we may have to adjust any OP_RECURSE calls inside the group. */ else { int offset; *code = OP_END; adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd); memmove(previous + 2 + LINK_SIZE, previous, len); code += 2 + LINK_SIZE; *previous++ = OP_BRAZERO + repeat_type; *previous++ = OP_BRA; /* We chain together the bracket offset fields that have to be filled in later when the ends of the brackets are reached. */ offset = (bralink == NULL)? 0 : previous - bralink; bralink = previous; PUTINC(previous, 0, offset); } repeat_max--; } /* If the minimum is greater than zero, replicate the group as many times as necessary, and adjust the maximum to the number of subsequent copies that we need. If we set a first char from the group, and didn't set a required char, copy the latter from the former. */ else { if (repeat_min > 1) { if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte; for (i = 1; i < repeat_min; i++) { memcpy(code, previous, len); code += len; } } if (repeat_max > 0) repeat_max -= repeat_min; } /* This code is common to both the zero and non-zero minimum cases. If the maximum is limited, it replicates the group in a nested fashion, remembering the bracket starts on a stack. In the case of a zero minimum, the first one was set up above. In all cases the repeat_max now specifies the number of additional copies needed. */ if (repeat_max >= 0) { for (i = repeat_max - 1; i >= 0; i--) { *code++ = OP_BRAZERO + repeat_type; /* All but the final copy start a new nesting, maintaining the chain of brackets outstanding. */ if (i != 0) { int offset; *code++ = OP_BRA; offset = (bralink == NULL)? 0 : code - bralink; bralink = code; PUTINC(code, 0, offset); } memcpy(code, previous, len); code += len; } /* Now chain through the pending brackets, and fill in their length fields (which are holding the chain links pro tem). */ while (bralink != NULL) { int oldlinkoffset; int offset = code - bralink + 1; uschar *bra = code - offset; oldlinkoffset = GET(bra, 1); bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset; *code++ = OP_KET; PUTINC(code, 0, offset); PUT(bra, 1, offset); } } /* If the maximum is unlimited, set a repeater in the final copy. We can't just offset backwards from the current code point, because we don't know if there's been an options resetting after the ket. The correct offset was computed above. */ else code[-ketoffset] = OP_KETRMAX + repeat_type; } /* Else there's some kind of shambles */ else { *errorptr = ERR11; goto FAILED; } /* If the character following a repeat is '+', we wrap the entire repeated item inside OP_ONCE brackets. This is just syntactic sugar, taken from Sun's Java package. The repeated item starts at tempcode, not at previous, which might be the first part of a string whose (former) last char we repeated. However, we don't support '+' after a greediness '?'. */ if (possessive_quantifier) { int len = code - tempcode; memmove(tempcode + 1+LINK_SIZE, tempcode, len); code += 1 + LINK_SIZE; len += 1 + LINK_SIZE; tempcode[0] = OP_ONCE; *code++ = OP_KET; PUTINC(code, 0, len); PUT(tempcode, 1, len); } /* In all case we no longer have a previous item. We also set the "follows varying string" flag for subsequently encountered reqbytes if it isn't already set and we have just passed a varying length item. */ END_REPEAT: previous = NULL; cd->req_varyopt |= reqvary; break; /* Start of nested bracket sub-expression, or comment or lookahead or lookbehind or option setting or condition. First deal with special things that can come after a bracket; all are introduced by ?, and the appearance of any of them means that this is not a referencing group. They were checked for validity in the first pass over the string, so we don't have to check for syntax errors here. */ case '(': newoptions = options; skipbytes = 0; if (*(++ptr) == '?') { int set, unset; int *optset; switch (*(++ptr)) { case '#': /* Comment; skip to ket */ ptr++; while (*ptr != ')') ptr++; continue; case ':': /* Non-extracting bracket */ bravalue = OP_BRA; ptr++; break; case '(': bravalue = OP_COND; /* Conditional group */ /* Condition to test for recursion */ if (ptr[1] == 'R') { code[1+LINK_SIZE] = OP_CREF; PUT2(code, 2+LINK_SIZE, CREF_RECURSE); skipbytes = 3; ptr += 3; } /* Condition to test for a numbered subpattern match. We know that if a digit follows ( then there will just be digits until ) because the syntax was checked in the first pass. */ else if ((digitab[ptr[1]] && ctype_digit) != 0) { int condref; /* Don't amalgamate; some compilers */ condref = *(++ptr) - '0'; /* grumble at autoincrement in declaration */ while (*(++ptr) != ')') condref = condref*10 + *ptr - '0'; if (condref == 0) { *errorptr = ERR35; goto FAILED; } ptr++; code[1+LINK_SIZE] = OP_CREF; PUT2(code, 2+LINK_SIZE, condref); skipbytes = 3; } /* For conditions that are assertions, we just fall through, having set bravalue above. */ break; case '=': /* Positive lookahead */ bravalue = OP_ASSERT; ptr++; break; case '!': /* Negative lookahead */ bravalue = OP_ASSERT_NOT; ptr++; break; case '<': /* Lookbehinds */ switch (*(++ptr)) { case '=': /* Positive lookbehind */ bravalue = OP_ASSERTBACK; ptr++; break; case '!': /* Negative lookbehind */ bravalue = OP_ASSERTBACK_NOT; ptr++; break; } break; case '>': /* One-time brackets */ bravalue = OP_ONCE; ptr++; break; case 'C': /* Callout - may be followed by digits; */ previous_callout = code; /* Save for later completion */ after_manual_callout = 1; /* Skip one item before completing */ *code++ = OP_CALLOUT; /* Already checked that the terminating */ { /* closing parenthesis is present. */ int n = 0; while ((digitab[*(++ptr)] & ctype_digit) != 0) n = n * 10 + *ptr - '0'; if (n > 255) { *errorptr = ERR38; goto FAILED; } *code++ = n; PUT(code, 0, ptr - cd->start_pattern + 1); /* Pattern offset */ PUT(code, LINK_SIZE, 0); /* Default length */ code += 2 * LINK_SIZE; } previous = NULL; continue; case 'P': /* Named subpattern handling */ if (*(++ptr) == '<') /* Definition */ { int i, namelen; uschar *slot = cd->name_table; const uschar *name; /* Don't amalgamate; some compilers */ name = ++ptr; /* grumble at autoincrement in declaration */ while (*ptr++ != '>'); namelen = ptr - name - 1; for (i = 0; i < cd->names_found; i++) { int crc = memcmp(name, slot+2, namelen); if (crc == 0) { if (slot[2+namelen] == 0) { *errorptr = ERR43; goto FAILED; } crc = -1; /* Current name is substring */ } if (crc < 0) { memmove(slot + cd->name_entry_size, slot, (cd->names_found - i) * cd->name_entry_size); break; } slot += cd->name_entry_size; } PUT2(slot, 0, *brackets + 1); memcpy(slot + 2, name, namelen); slot[2+namelen] = 0; cd->names_found++; goto NUMBERED_GROUP; } if (*ptr == '=' || *ptr == '>') /* Reference or recursion */ { int i, namelen; int type = *ptr++; const uschar *name = ptr; uschar *slot = cd->name_table; while (*ptr != ')') ptr++; namelen = ptr - name; for (i = 0; i < cd->names_found; i++) { if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break; slot += cd->name_entry_size; } if (i >= cd->names_found) { *errorptr = ERR15; goto FAILED; } recno = GET2(slot, 0); if (type == '>') goto HANDLE_RECURSION; /* A few lines below */ /* Back reference */ previous = code; *code++ = OP_REF; PUT2INC(code, 0, recno); cd->backref_map |= (recno < 32)? (1 << recno) : 1; if (recno > cd->top_backref) cd->top_backref = recno; continue; } /* Should never happen */ break; case 'R': /* Pattern recursion */ ptr++; /* Same as (?0) */ /* Fall through */ /* Recursion or "subroutine" call */ case '0': case '1': case '2': case '3': case '4': case '5': case '6': case '7': case '8': case '9': { const uschar *called; recno = 0; while((digitab[*ptr] & ctype_digit) != 0) recno = recno * 10 + *ptr++ - '0'; /* Come here from code above that handles a named recursion */ HANDLE_RECURSION: previous = code; /* Find the bracket that is being referenced. Temporarily end the regex in case it doesn't exist. */ *code = OP_END; called = (recno == 0)? cd->start_code : find_bracket(cd->start_code, utf8, recno); if (called == NULL) { *errorptr = ERR15; goto FAILED; } /* If the subpattern is still open, this is a recursive call. We check to see if this is a left recursion that could loop for ever, and diagnose that case. */ if (GET(called, 1) == 0 && could_be_empty(called, code, bcptr, utf8)) { *errorptr = ERR40; goto FAILED; } /* Insert the recursion/subroutine item */ *code = OP_RECURSE; PUT(code, 1, called - cd->start_code); code += 1 + LINK_SIZE; } continue; /* Character after (? not specially recognized */ default: /* Option setting */ set = unset = 0; optset = &set; while (*ptr != ')' && *ptr != ':') { switch (*ptr++) { case '-': optset = &unset; break; case 'i': *optset |= PCRE_CASELESS; break; case 'm': *optset |= PCRE_MULTILINE; break; case 's': *optset |= PCRE_DOTALL; break; case 'x': *optset |= PCRE_EXTENDED; break; case 'U': *optset |= PCRE_UNGREEDY; break; case 'X': *optset |= PCRE_EXTRA; break; } } /* Set up the changed option bits, but don't change anything yet. */ newoptions = (options | set) & (~unset); /* If the options ended with ')' this is not the start of a nested group with option changes, so the options change at this level. Compile code to change the ims options if this setting actually changes any of them. We also pass the new setting back so that it can be put at the start of any following branches, and when this group ends (if we are in a group), a resetting item can be compiled. Note that if this item is right at the start of the pattern, the options will have been abstracted and made global, so there will be no change to compile. */ if (*ptr == ')') { if ((options & PCRE_IMS) != (newoptions & PCRE_IMS)) { *code++ = OP_OPT; *code++ = newoptions & PCRE_IMS; } /* Change options at this level, and pass them back for use in subsequent branches. Reset the greedy defaults and the case value for firstbyte and reqbyte. */ *optionsptr = options = newoptions; greedy_default = ((newoptions & PCRE_UNGREEDY) != 0); greedy_non_default = greedy_default ^ 1; req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0; previous = NULL; /* This item can't be repeated */ continue; /* It is complete */ } /* If the options ended with ':' we are heading into a nested group with possible change of options. Such groups are non-capturing and are not assertions of any kind. All we need to do is skip over the ':'; the newoptions value is handled below. */ bravalue = OP_BRA; ptr++; } } /* If PCRE_NO_AUTO_CAPTURE is set, all unadorned brackets become non-capturing and behave like (?:...) brackets */ else if ((options & PCRE_NO_AUTO_CAPTURE) != 0) { bravalue = OP_BRA; } /* Else we have a referencing group; adjust the opcode. If the bracket number is greater than EXTRACT_BASIC_MAX, we set the opcode one higher, and arrange for the true number to follow later, in an OP_BRANUMBER item. */ else { NUMBERED_GROUP: if (++(*brackets) > EXTRACT_BASIC_MAX) { bravalue = OP_BRA + EXTRACT_BASIC_MAX + 1; code[1+LINK_SIZE] = OP_BRANUMBER; PUT2(code, 2+LINK_SIZE, *brackets); skipbytes = 3; } else bravalue = OP_BRA + *brackets; } /* Process nested bracketed re. Assertions may not be repeated, but other kinds can be. We copy code into a non-register variable in order to be able to pass its address because some compilers complain otherwise. Pass in a new setting for the ims options if they have changed. */ previous = (bravalue >= OP_ONCE)? code : NULL; *code = bravalue; tempcode = code; tempreqvary = cd->req_varyopt; /* Save value before bracket */ if (!compile_regex( newoptions, /* The complete new option state */ options & PCRE_IMS, /* The previous ims option state */ brackets, /* Extracting bracket count */ &tempcode, /* Where to put code (updated) */ &ptr, /* Input pointer (updated) */ errorptr, /* Where to put an error message */ (bravalue == OP_ASSERTBACK || bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */ skipbytes, /* Skip over OP_COND/OP_BRANUMBER */ &subfirstbyte, /* For possible first char */ &subreqbyte, /* For possible last char */ bcptr, /* Current branch chain */ cd)) /* Tables block */ goto FAILED; /* At the end of compiling, code is still pointing to the start of the group, while tempcode has been updated to point past the end of the group and any option resetting that may follow it. The pattern pointer (ptr) is on the bracket. */ /* If this is a conditional bracket, check that there are no more than two branches in the group. */ else if (bravalue == OP_COND) { uschar *tc = code; condcount = 0; do { condcount++; tc += GET(tc,1); } while (*tc != OP_KET); if (condcount > 2) { *errorptr = ERR27; goto FAILED; } /* If there is just one branch, we must not make use of its firstbyte or reqbyte, because this is equivalent to an empty second branch. */ if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE; } /* Handle updating of the required and first characters. Update for normal brackets of all kinds, and conditions with two branches (see code above). If the bracket is followed by a quantifier with zero repeat, we have to back off. Hence the definition of zeroreqbyte and zerofirstbyte outside the main loop so that they can be accessed for the back off. */ zeroreqbyte = reqbyte; zerofirstbyte = firstbyte; groupsetfirstbyte = FALSE; if (bravalue >= OP_BRA || bravalue == OP_ONCE || bravalue == OP_COND) { /* If we have not yet set a firstbyte in this branch, take it from the subpattern, remembering that it was set here so that a repeat of more than one can replicate it as reqbyte if necessary. If the subpattern has no firstbyte, set "none" for the whole branch. In both cases, a zero repeat forces firstbyte to "none". */ if (firstbyte == REQ_UNSET) { if (subfirstbyte >= 0) { firstbyte = subfirstbyte; groupsetfirstbyte = TRUE; } else firstbyte = REQ_NONE; zerofirstbyte = REQ_NONE; } /* If firstbyte was previously set, convert the subpattern's firstbyte into reqbyte if there wasn't one, using the vary flag that was in existence beforehand. */ else if (subfirstbyte >= 0 && subreqbyte < 0) subreqbyte = subfirstbyte | tempreqvary; /* If the subpattern set a required byte (or set a first byte that isn't really the first byte - see above), set it. */ if (subreqbyte >= 0) reqbyte = subreqbyte; } /* For a forward assertion, we take the reqbyte, if set. This can be helpful if the pattern that follows the assertion doesn't set a different char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte for an assertion, however because it leads to incorrect effect for patterns such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead of a firstbyte. This is overcome by a scan at the end if there's no firstbyte, looking for an asserted first char. */ else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte; /* Now update the main code pointer to the end of the group. */ code = tempcode; /* Error if hit end of pattern */ if (*ptr != ')') { *errorptr = ERR14; goto FAILED; } break; /* Check \ for being a real metacharacter; if not, fall through and handle it as a data character at the start of a string. Escape items are checked for validity in the pre-compiling pass. */ case '\\': tempptr = ptr; c = check_escape(&ptr, errorptr, *brackets, options, FALSE); /* Handle metacharacters introduced by \. For ones like \d, the ESC_ values are arranged to be the negation of the corresponding OP_values. For the back references, the values are ESC_REF plus the reference number. Only back references and those types that consume a character may be repeated. We can test for values between ESC_b and ESC_Z for the latter; this may have to change if any new ones are ever created. */ if (c < 0) { if (-c == ESC_Q) /* Handle start of quoted string */ { if (ptr[1] == '\\' && ptr[2] == 'E') ptr += 2; /* avoid empty string */ else inescq = TRUE; continue; } /* For metasequences that actually match a character, we disable the setting of a first character if it hasn't already been set. */ if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z) firstbyte = REQ_NONE; /* Set values to reset to if this is followed by a zero repeat. */ zerofirstbyte = firstbyte; zeroreqbyte = reqbyte; /* Back references are handled specially */ if (-c >= ESC_REF) { int number = -c - ESC_REF; previous = code; *code++ = OP_REF; PUT2INC(code, 0, number); } /* So are Unicode property matches, if supported. We know that get_ucp won't fail because it was tested in the pre-pass. */ #ifdef SUPPORT_UCP else if (-c == ESC_P || -c == ESC_p) { PCRE_BOOL negated; int value = get_ucp(&ptr, &negated, errorptr); previous = code; *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP; *code++ = value; } #endif /* For the rest, we can obtain the OP value by negating the escape value */ else { previous = (-c > ESC_b && -c < ESC_Z)? code : NULL; *code++ = -c; } continue; } /* We have a data character whose value is in c. In UTF-8 mode it may have a value > 127. We set its representation in the length/buffer, and then handle it as a data character. */ #ifdef SUPPORT_UTF8 if (utf8 && c > 127) mclength = ord2utf8(c, mcbuffer); else #endif { mcbuffer[0] = c; mclength = 1; } goto ONE_CHAR; /* Handle a literal character. It is guaranteed not to be whitespace or # when the extended flag is set. If we are in UTF-8 mode, it may be a multi-byte literal character. */ default: NORMAL_CHAR: mclength = 1; mcbuffer[0] = c; #ifdef SUPPORT_UTF8 if (utf8 && (c & 0xc0) == 0xc0) { while ((ptr[1] & 0xc0) == 0x80) mcbuffer[mclength++] = *(++ptr); } #endif /* At this point we have the character's bytes in mcbuffer, and the length in mclength. When not in UTF-8 mode, the length is always 1. */ ONE_CHAR: previous = code; *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARNC : OP_CHAR; for (c = 0; c < mclength; c++) *code++ = mcbuffer[c]; /* Set the first and required bytes appropriately. If no previous first byte, set it from this character, but revert to none on a zero repeat. Otherwise, leave the firstbyte value alone, and don't change it on a zero repeat. */ if (firstbyte == REQ_UNSET) { zerofirstbyte = REQ_NONE; zeroreqbyte = reqbyte; /* If the character is more than one byte long, we can set firstbyte only if it is not to be matched caselessly. */ if (mclength == 1 || req_caseopt == 0) { firstbyte = mcbuffer[0] | req_caseopt; if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt; } else firstbyte = reqbyte = REQ_NONE; } /* firstbyte was previously set; we can set reqbyte only the length is 1 or the matching is caseful. */ else { zerofirstbyte = firstbyte; zeroreqbyte = reqbyte; if (mclength == 1 || req_caseopt == 0) reqbyte = code[-1] | req_caseopt | cd->req_varyopt; } break; /* End of literal character handling */ } } /* end of big loop */ /* Control never reaches here by falling through, only by a goto for all the error states. Pass back the position in the pattern so that it can be displayed to the user for diagnosing the error. */ FAILED: *ptrptr = ptr; return FALSE; } /************************************************* * Compile sequence of alternatives * *************************************************/ /* On entry, ptr is pointing past the bracket character, but on return it points to the closing bracket, or vertical bar, or end of string. The code variable is pointing at the byte into which the BRA operator has been stored. If the ims options are changed at the start (for a (?ims: group) or during any branch, we need to insert an OP_OPT item at the start of every following branch to ensure they get set correctly at run time, and also pass the new options into every subsequent branch compile. Argument: options option bits, including any changes for this subpattern oldims previous settings of ims option bits brackets -> int containing the number of extracting brackets used codeptr -> the address of the current code pointer ptrptr -> the address of the current pattern pointer errorptr -> pointer to error message lookbehind TRUE if this is a lookbehind assertion skipbytes skip this many bytes at start (for OP_COND, OP_BRANUMBER) firstbyteptr place to put the first required character, or a negative number reqbyteptr place to put the last required character, or a negative number bcptr pointer to the chain of currently open branches cd points to the data block with tables pointers etc. Returns: TRUE on success */ static PCRE_BOOL compile_regex(int options, int oldims, int *brackets, uschar **codeptr, const uschar **ptrptr, const char **errorptr, PCRE_BOOL lookbehind, int skipbytes, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd) { const uschar *ptr = *ptrptr; uschar *code = *codeptr; uschar *last_branch = code; uschar *start_bracket = code; uschar *reverse_count = NULL; int firstbyte, reqbyte; int branchfirstbyte, branchreqbyte; branch_chain bc; bc.outer = bcptr; bc.current = code; firstbyte = reqbyte = REQ_UNSET; /* Offset is set zero to mark that this bracket is still open */ PUT(code, 1, 0); code += 1 + LINK_SIZE + skipbytes; /* Loop for each alternative branch */ for (;;) { /* Handle a change of ims options at the start of the branch */ if ((options & PCRE_IMS) != oldims) { *code++ = OP_OPT; *code++ = options & PCRE_IMS; } /* Set up dummy OP_REVERSE if lookbehind assertion */ if (lookbehind) { *code++ = OP_REVERSE; reverse_count = code; PUTINC(code, 0, 0); } /* Now compile the branch */ if (!compile_branch(&options, brackets, &code, &ptr, errorptr, &branchfirstbyte, &branchreqbyte, &bc, cd)) { *ptrptr = ptr; return FALSE; } /* If this is the first branch, the firstbyte and reqbyte values for the branch become the values for the regex. */ if (*last_branch != OP_ALT) { firstbyte = branchfirstbyte; reqbyte = branchreqbyte; } /* If this is not the first branch, the first char and reqbyte have to match the values from all the previous branches, except that if the previous value for reqbyte didn't have REQ_VARY set, it can still match, and we set REQ_VARY for the regex. */ else { /* If we previously had a firstbyte, but it doesn't match the new branch, we have to abandon the firstbyte for the regex, but if there was previously no reqbyte, it takes on the value of the old firstbyte. */ if (firstbyte >= 0 && firstbyte != branchfirstbyte) { if (reqbyte < 0) reqbyte = firstbyte; firstbyte = REQ_NONE; } /* If we (now or from before) have no firstbyte, a firstbyte from the branch becomes a reqbyte if there isn't a branch reqbyte. */ if (firstbyte < 0 && branchfirstbyte >= 0 && branchreqbyte < 0) branchreqbyte = branchfirstbyte; /* Now ensure that the reqbytes match */ if ((reqbyte & ~REQ_VARY) != (branchreqbyte & ~REQ_VARY)) reqbyte = REQ_NONE; else reqbyte |= branchreqbyte; /* To "or" REQ_VARY */ } /* If lookbehind, check that this branch matches a fixed-length string, and put the length into the OP_REVERSE item. Temporarily mark the end of the branch with OP_END. */ if (lookbehind) { int length; *code = OP_END; length = find_fixedlength(last_branch, options); DPRINTF(("fixed length = %d\n", length)); if (length < 0) { *errorptr = (length == -2)? ERR36 : ERR25; *ptrptr = ptr; return FALSE; } PUT(reverse_count, 0, length); } /* Reached end of expression, either ')' or end of pattern. Go back through the alternative branches and reverse the chain of offsets, with the field in the BRA item now becoming an offset to the first alternative. If there are no alternatives, it points to the end of the group. The length in the terminating ket is always the length of the whole bracketed item. If any of the ims options were changed inside the group, compile a resetting op-code following, except at the very end of the pattern. Return leaving the pointer at the terminating char. */ if (*ptr != '|') { int length = code - last_branch; do { int prev_length = GET(last_branch, 1); PUT(last_branch, 1, length); length = prev_length; last_branch -= length; } while (length > 0); /* Fill in the ket */ *code = OP_KET; PUT(code, 1, code - start_bracket); code += 1 + LINK_SIZE; /* Resetting option if needed */ if ((options & PCRE_IMS) != oldims && *ptr == ')') { *code++ = OP_OPT; *code++ = oldims; } /* Set values to pass back */ *codeptr = code; *ptrptr = ptr; *firstbyteptr = firstbyte; *reqbyteptr = reqbyte; return TRUE; } /* Another branch follows; insert an "or" node. Its length field points back to the previous branch while the bracket remains open. At the end the chain is reversed. It's done like this so that the start of the bracket has a zero offset until it is closed, making it possible to detect recursion. */ *code = OP_ALT; PUT(code, 1, code - last_branch); bc.current = last_branch = code; code += 1 + LINK_SIZE; ptr++; } /* Control never reaches here */ } /************************************************* * Check for anchored expression * *************************************************/ /* Try to find out if this is an anchored regular expression. Consider each alternative branch. If they all start with OP_SOD or OP_CIRC, or with a bracket all of whose alternatives start with OP_SOD or OP_CIRC (recurse ad lib), then it's anchored. However, if this is a multiline pattern, then only OP_SOD counts, since OP_CIRC can match in the middle. We can also consider a regex to be anchored if OP_SOM starts all its branches. This is the code for \G, which means "match at start of match position, taking into account the match offset". A branch is also implicitly anchored if it starts with .* and DOTALL is set, because that will try the rest of the pattern at all possible matching points, so there is no point trying again.... er .... .... except when the .* appears inside capturing parentheses, and there is a subsequent back reference to those parentheses. We haven't enough information to catch that case precisely. At first, the best we could do was to detect when .* was in capturing brackets and the highest back reference was greater than or equal to that level. However, by keeping a bitmap of the first 31 back references, we can catch some of the more common cases more precisely. Arguments: code points to start of expression (the bracket) opti