]>
Commit | Line | Data |
---|---|---|
9d3616dc AF |
1 | diff -Naur linux-3.14.22.org/include/linux/netfilter/xt_layer7.h linux-3.14.22/include/linux/netfilter/xt_layer7.h |
2 | --- linux-3.14.22.org/include/linux/netfilter/xt_layer7.h 1970-01-01 01:00:00.000000000 +0100 | |
3 | +++ linux-3.14.22/include/linux/netfilter/xt_layer7.h 2014-10-24 15:33:21.281274380 +0200 | |
4 | @@ -0,0 +1,14 @@ | |
5 | +#ifndef _XT_LAYER7_H | |
6 | +#define _XT_LAYER7_H | |
7 | + | |
8 | +#define MAX_PATTERN_LEN 8192 | |
9 | +#define MAX_PROTOCOL_LEN 256 | |
10 | + | |
11 | +struct xt_layer7_info { | |
12 | + char protocol[MAX_PROTOCOL_LEN]; | |
13 | + char pattern[MAX_PATTERN_LEN]; | |
14 | + u_int8_t invert; | |
15 | + u_int8_t pkt; | |
16 | +}; | |
17 | + | |
18 | +#endif /* _XT_LAYER7_H */ | |
19 | diff -Naur linux-3.14.22.org/include/net/netfilter/nf_conntrack.h linux-3.14.22/include/net/netfilter/nf_conntrack.h | |
20 | --- linux-3.14.22.org/include/net/netfilter/nf_conntrack.h 2014-10-15 08:42:04.000000000 +0200 | |
21 | +++ linux-3.14.22/include/net/netfilter/nf_conntrack.h 2014-10-24 15:33:21.281274380 +0200 | |
22 | @@ -105,6 +105,22 @@ | |
23 | struct net *ct_net; | |
24 | #endif | |
25 | ||
26 | +#if defined(CONFIG_NETFILTER_XT_MATCH_LAYER7) || \ | |
27 | + defined(CONFIG_NETFILTER_XT_MATCH_LAYER7_MODULE) | |
28 | + struct { | |
29 | + /* | |
30 | + * e.g. "http". NULL before decision. "unknown" after decision | |
31 | + * if no match. | |
32 | + */ | |
33 | + char *app_proto; | |
34 | + /* | |
35 | + * application layer data so far. NULL after match decision. | |
36 | + */ | |
37 | + char *app_data; | |
38 | + unsigned int app_data_len; | |
39 | + } layer7; | |
40 | +#endif | |
41 | + | |
42 | /* Storage reserved for other modules, must be the last member */ | |
43 | union nf_conntrack_proto proto; | |
44 | }; | |
45 | diff -Naur linux-3.14.22.org/include/uapi/linux/netfilter/Kbuild linux-3.14.22/include/uapi/linux/netfilter/Kbuild | |
46 | --- linux-3.14.22.org/include/uapi/linux/netfilter/Kbuild 2014-10-15 08:42:04.000000000 +0200 | |
47 | +++ linux-3.14.22/include/uapi/linux/netfilter/Kbuild 2014-10-24 15:34:39.721663344 +0200 | |
48 | @@ -59,6 +59,7 @@ | |
49 | header-y += xt_iprange.h | |
50 | header-y += xt_ipvs.h | |
51 | header-y += xt_l2tp.h | |
52 | +header-y += xt_layer7.h | |
53 | header-y += xt_length.h | |
54 | header-y += xt_limit.h | |
55 | header-y += xt_mac.h | |
56 | diff -Naur linux-3.14.22.org/net/netfilter/Kconfig linux-3.14.22/net/netfilter/Kconfig | |
57 | --- linux-3.14.22.org/net/netfilter/Kconfig 2014-10-15 08:42:04.000000000 +0200 | |
58 | +++ linux-3.14.22/net/netfilter/Kconfig 2014-10-24 15:33:21.281274380 +0200 | |
59 | @@ -1153,6 +1153,26 @@ | |
60 | ||
61 | To compile it as a module, choose M here. If unsure, say N. | |
62 | ||
63 | +config NETFILTER_XT_MATCH_LAYER7 | |
64 | + tristate '"layer7" match support' | |
65 | + depends on NETFILTER_XTABLES | |
66 | + depends on NETFILTER_ADVANCED | |
67 | + depends on NF_CONNTRACK | |
68 | + help | |
69 | + Say Y if you want to be able to classify connections (and their | |
70 | + packets) based on regular expression matching of their application | |
71 | + layer data. This is one way to classify applications such as | |
72 | + peer-to-peer filesharing systems that do not always use the same | |
73 | + port. | |
74 | + | |
75 | + To compile it as a module, choose M here. If unsure, say N. | |
76 | + | |
77 | +config NETFILTER_XT_MATCH_LAYER7_DEBUG | |
78 | + bool 'Layer 7 debugging output' | |
79 | + depends on NETFILTER_XT_MATCH_LAYER7 | |
80 | + help | |
81 | + Say Y to get lots of debugging output. | |
82 | + | |
83 | config NETFILTER_XT_MATCH_LENGTH | |
84 | tristate '"length" match support' | |
85 | depends on NETFILTER_ADVANCED | |
86 | @@ -1347,6 +1367,12 @@ | |
87 | ||
88 | To compile it as a module, choose M here. If unsure, say N. | |
89 | ||
90 | +config NETFILTER_XT_MATCH_LAYER7_DEBUG | |
91 | + bool 'Layer 7 debugging output' | |
92 | + depends on NETFILTER_XT_MATCH_LAYER7 | |
93 | + help | |
94 | + Say Y to get lots of debugging output. | |
95 | + | |
96 | config NETFILTER_XT_MATCH_STATISTIC | |
97 | tristate '"statistic" match support' | |
98 | depends on NETFILTER_ADVANCED | |
99 | diff -Naur linux-3.14.22.org/net/netfilter/Makefile linux-3.14.22/net/netfilter/Makefile | |
100 | --- linux-3.14.22.org/net/netfilter/Makefile 2014-10-15 08:42:04.000000000 +0200 | |
101 | +++ linux-3.14.22/net/netfilter/Makefile 2014-10-24 15:33:21.281274380 +0200 | |
102 | @@ -158,6 +158,7 @@ | |
103 | obj-$(CONFIG_NETFILTER_XT_MATCH_SCTP) += xt_sctp.o | |
104 | obj-$(CONFIG_NETFILTER_XT_MATCH_SOCKET) += xt_socket.o | |
105 | obj-$(CONFIG_NETFILTER_XT_MATCH_STATE) += xt_state.o | |
106 | +obj-$(CONFIG_NETFILTER_XT_MATCH_LAYER7) += xt_layer7.o | |
107 | obj-$(CONFIG_NETFILTER_XT_MATCH_STATISTIC) += xt_statistic.o | |
108 | obj-$(CONFIG_NETFILTER_XT_MATCH_STRING) += xt_string.o | |
109 | obj-$(CONFIG_NETFILTER_XT_MATCH_TCPMSS) += xt_tcpmss.o | |
110 | diff -Naur linux-3.14.22.org/net/netfilter/nf_conntrack_core.c linux-3.14.22/net/netfilter/nf_conntrack_core.c | |
111 | --- linux-3.14.22.org/net/netfilter/nf_conntrack_core.c 2014-10-15 08:42:04.000000000 +0200 | |
112 | +++ linux-3.14.22/net/netfilter/nf_conntrack_core.c 2014-10-24 15:33:21.281274380 +0200 | |
113 | @@ -220,6 +220,13 @@ | |
114 | * too. */ | |
115 | nf_ct_remove_expectations(ct); | |
116 | ||
117 | +#if defined(CONFIG_NETFILTER_XT_MATCH_LAYER7) || defined(CONFIG_NETFILTER_XT_MATCH_LAYER7_MODULE) | |
118 | + if(ct->layer7.app_proto) | |
119 | + kfree(ct->layer7.app_proto); | |
120 | + if(ct->layer7.app_data) | |
121 | + kfree(ct->layer7.app_data); | |
122 | +#endif | |
123 | + | |
124 | /* We overload first tuple to link into unconfirmed or dying list.*/ | |
125 | BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode)); | |
126 | hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); | |
127 | diff -Naur linux-3.14.22.org/net/netfilter/nf_conntrack_standalone.c linux-3.14.22/net/netfilter/nf_conntrack_standalone.c | |
128 | --- linux-3.14.22.org/net/netfilter/nf_conntrack_standalone.c 2014-10-15 08:42:04.000000000 +0200 | |
129 | +++ linux-3.14.22/net/netfilter/nf_conntrack_standalone.c 2014-10-24 15:33:21.285274399 +0200 | |
130 | @@ -240,6 +240,12 @@ | |
131 | if (ct_show_delta_time(s, ct)) | |
132 | goto release; | |
133 | ||
134 | +#if defined(CONFIG_NETFILTER_XT_MATCH_LAYER7) || defined(CONFIG_NETFILTER_XT_MATCH_LAYER7_MODULE) | |
135 | + if(ct->layer7.app_proto && | |
136 | + seq_printf(s, "l7proto=%s ", ct->layer7.app_proto)) | |
137 | + return -ENOSPC; | |
138 | +#endif | |
139 | + | |
140 | if (seq_printf(s, "use=%u\n", atomic_read(&ct->ct_general.use))) | |
141 | goto release; | |
142 | ||
143 | diff -Naur linux-3.14.22.org/net/netfilter/regexp/regexp.c linux-3.14.22/net/netfilter/regexp/regexp.c | |
144 | --- linux-3.14.22.org/net/netfilter/regexp/regexp.c 1970-01-01 01:00:00.000000000 +0100 | |
145 | +++ linux-3.14.22/net/netfilter/regexp/regexp.c 2014-10-24 15:33:21.285274399 +0200 | |
146 | @@ -0,0 +1,1197 @@ | |
147 | +/* | |
148 | + * regcomp and regexec -- regsub and regerror are elsewhere | |
149 | + * @(#)regexp.c 1.3 of 18 April 87 | |
150 | + * | |
151 | + * Copyright (c) 1986 by University of Toronto. | |
152 | + * Written by Henry Spencer. Not derived from licensed software. | |
153 | + * | |
154 | + * Permission is granted to anyone to use this software for any | |
155 | + * purpose on any computer system, and to redistribute it freely, | |
156 | + * subject to the following restrictions: | |
157 | + * | |
158 | + * 1. The author is not responsible for the consequences of use of | |
159 | + * this software, no matter how awful, even if they arise | |
160 | + * from defects in it. | |
161 | + * | |
162 | + * 2. The origin of this software must not be misrepresented, either | |
163 | + * by explicit claim or by omission. | |
164 | + * | |
165 | + * 3. Altered versions must be plainly marked as such, and must not | |
166 | + * be misrepresented as being the original software. | |
167 | + * | |
168 | + * Beware that some of this code is subtly aware of the way operator | |
169 | + * precedence is structured in regular expressions. Serious changes in | |
170 | + * regular-expression syntax might require a total rethink. | |
171 | + * | |
172 | + * This code was modified by Ethan Sommer to work within the kernel | |
173 | + * (it now uses kmalloc etc..) | |
174 | + * | |
175 | + * Modified slightly by Matthew Strait to use more modern C. | |
176 | + */ | |
177 | + | |
178 | +#include "regexp.h" | |
179 | +#include "regmagic.h" | |
180 | + | |
181 | +/* added by ethan and matt. Lets it work in both kernel and user space. | |
182 | +(So iptables can use it, for instance.) Yea, it goes both ways... */ | |
183 | +#if __KERNEL__ | |
184 | + #define malloc(foo) kmalloc(foo,GFP_ATOMIC) | |
185 | +#else | |
186 | + #define printk(format,args...) printf(format,##args) | |
187 | +#endif | |
188 | + | |
189 | +void regerror(char * s) | |
190 | +{ | |
191 | + printk("<3>Regexp: %s\n", s); | |
192 | + /* NOTREACHED */ | |
193 | +} | |
194 | + | |
195 | +/* | |
196 | + * The "internal use only" fields in regexp.h are present to pass info from | |
197 | + * compile to execute that permits the execute phase to run lots faster on | |
198 | + * simple cases. They are: | |
199 | + * | |
200 | + * regstart char that must begin a match; '\0' if none obvious | |
201 | + * reganch is the match anchored (at beginning-of-line only)? | |
202 | + * regmust string (pointer into program) that match must include, or NULL | |
203 | + * regmlen length of regmust string | |
204 | + * | |
205 | + * Regstart and reganch permit very fast decisions on suitable starting points | |
206 | + * for a match, cutting down the work a lot. Regmust permits fast rejection | |
207 | + * of lines that cannot possibly match. The regmust tests are costly enough | |
208 | + * that regcomp() supplies a regmust only if the r.e. contains something | |
209 | + * potentially expensive (at present, the only such thing detected is * or + | |
210 | + * at the start of the r.e., which can involve a lot of backup). Regmlen is | |
211 | + * supplied because the test in regexec() needs it and regcomp() is computing | |
212 | + * it anyway. | |
213 | + */ | |
214 | + | |
215 | +/* | |
216 | + * Structure for regexp "program". This is essentially a linear encoding | |
217 | + * of a nondeterministic finite-state machine (aka syntax charts or | |
218 | + * "railroad normal form" in parsing technology). Each node is an opcode | |
219 | + * plus a "next" pointer, possibly plus an operand. "Next" pointers of | |
220 | + * all nodes except BRANCH implement concatenation; a "next" pointer with | |
221 | + * a BRANCH on both ends of it is connecting two alternatives. (Here we | |
222 | + * have one of the subtle syntax dependencies: an individual BRANCH (as | |
223 | + * opposed to a collection of them) is never concatenated with anything | |
224 | + * because of operator precedence.) The operand of some types of node is | |
225 | + * a literal string; for others, it is a node leading into a sub-FSM. In | |
226 | + * particular, the operand of a BRANCH node is the first node of the branch. | |
227 | + * (NB this is *not* a tree structure: the tail of the branch connects | |
228 | + * to the thing following the set of BRANCHes.) The opcodes are: | |
229 | + */ | |
230 | + | |
231 | +/* definition number opnd? meaning */ | |
232 | +#define END 0 /* no End of program. */ | |
233 | +#define BOL 1 /* no Match "" at beginning of line. */ | |
234 | +#define EOL 2 /* no Match "" at end of line. */ | |
235 | +#define ANY 3 /* no Match any one character. */ | |
236 | +#define ANYOF 4 /* str Match any character in this string. */ | |
237 | +#define ANYBUT 5 /* str Match any character not in this string. */ | |
238 | +#define BRANCH 6 /* node Match this alternative, or the next... */ | |
239 | +#define BACK 7 /* no Match "", "next" ptr points backward. */ | |
240 | +#define EXACTLY 8 /* str Match this string. */ | |
241 | +#define NOTHING 9 /* no Match empty string. */ | |
242 | +#define STAR 10 /* node Match this (simple) thing 0 or more times. */ | |
243 | +#define PLUS 11 /* node Match this (simple) thing 1 or more times. */ | |
244 | +#define OPEN 20 /* no Mark this point in input as start of #n. */ | |
245 | + /* OPEN+1 is number 1, etc. */ | |
246 | +#define CLOSE 30 /* no Analogous to OPEN. */ | |
247 | + | |
248 | +/* | |
249 | + * Opcode notes: | |
250 | + * | |
251 | + * BRANCH The set of branches constituting a single choice are hooked | |
252 | + * together with their "next" pointers, since precedence prevents | |
253 | + * anything being concatenated to any individual branch. The | |
254 | + * "next" pointer of the last BRANCH in a choice points to the | |
255 | + * thing following the whole choice. This is also where the | |
256 | + * final "next" pointer of each individual branch points; each | |
257 | + * branch starts with the operand node of a BRANCH node. | |
258 | + * | |
259 | + * BACK Normal "next" pointers all implicitly point forward; BACK | |
260 | + * exists to make loop structures possible. | |
261 | + * | |
262 | + * STAR,PLUS '?', and complex '*' and '+', are implemented as circular | |
263 | + * BRANCH structures using BACK. Simple cases (one character | |
264 | + * per match) are implemented with STAR and PLUS for speed | |
265 | + * and to minimize recursive plunges. | |
266 | + * | |
267 | + * OPEN,CLOSE ...are numbered at compile time. | |
268 | + */ | |
269 | + | |
270 | +/* | |
271 | + * A node is one char of opcode followed by two chars of "next" pointer. | |
272 | + * "Next" pointers are stored as two 8-bit pieces, high order first. The | |
273 | + * value is a positive offset from the opcode of the node containing it. | |
274 | + * An operand, if any, simply follows the node. (Note that much of the | |
275 | + * code generation knows about this implicit relationship.) | |
276 | + * | |
277 | + * Using two bytes for the "next" pointer is vast overkill for most things, | |
278 | + * but allows patterns to get big without disasters. | |
279 | + */ | |
280 | +#define OP(p) (*(p)) | |
281 | +#define NEXT(p) (((*((p)+1)&0377)<<8) + (*((p)+2)&0377)) | |
282 | +#define OPERAND(p) ((p) + 3) | |
283 | + | |
284 | +/* | |
285 | + * See regmagic.h for one further detail of program structure. | |
286 | + */ | |
287 | + | |
288 | + | |
289 | +/* | |
290 | + * Utility definitions. | |
291 | + */ | |
292 | +#ifndef CHARBITS | |
293 | +#define UCHARAT(p) ((int)*(unsigned char *)(p)) | |
294 | +#else | |
295 | +#define UCHARAT(p) ((int)*(p)&CHARBITS) | |
296 | +#endif | |
297 | + | |
298 | +#define FAIL(m) { regerror(m); return(NULL); } | |
299 | +#define ISMULT(c) ((c) == '*' || (c) == '+' || (c) == '?') | |
300 | +#define META "^$.[()|?+*\\" | |
301 | + | |
302 | +/* | |
303 | + * Flags to be passed up and down. | |
304 | + */ | |
305 | +#define HASWIDTH 01 /* Known never to match null string. */ | |
306 | +#define SIMPLE 02 /* Simple enough to be STAR/PLUS operand. */ | |
307 | +#define SPSTART 04 /* Starts with * or +. */ | |
308 | +#define WORST 0 /* Worst case. */ | |
309 | + | |
310 | +/* | |
311 | + * Global work variables for regcomp(). | |
312 | + */ | |
313 | +struct match_globals { | |
314 | +char *reginput; /* String-input pointer. */ | |
315 | +char *regbol; /* Beginning of input, for ^ check. */ | |
316 | +char **regstartp; /* Pointer to startp array. */ | |
317 | +char **regendp; /* Ditto for endp. */ | |
318 | +char *regparse; /* Input-scan pointer. */ | |
319 | +int regnpar; /* () count. */ | |
320 | +char regdummy; | |
321 | +char *regcode; /* Code-emit pointer; ®dummy = don't. */ | |
322 | +long regsize; /* Code size. */ | |
323 | +}; | |
324 | + | |
325 | +/* | |
326 | + * Forward declarations for regcomp()'s friends. | |
327 | + */ | |
328 | +#ifndef STATIC | |
329 | +#define STATIC static | |
330 | +#endif | |
331 | +STATIC char *reg(struct match_globals *g, int paren,int *flagp); | |
332 | +STATIC char *regbranch(struct match_globals *g, int *flagp); | |
333 | +STATIC char *regpiece(struct match_globals *g, int *flagp); | |
334 | +STATIC char *regatom(struct match_globals *g, int *flagp); | |
335 | +STATIC char *regnode(struct match_globals *g, char op); | |
336 | +STATIC char *regnext(struct match_globals *g, char *p); | |
337 | +STATIC void regc(struct match_globals *g, char b); | |
338 | +STATIC void reginsert(struct match_globals *g, char op, char *opnd); | |
339 | +STATIC void regtail(struct match_globals *g, char *p, char *val); | |
340 | +STATIC void regoptail(struct match_globals *g, char *p, char *val); | |
341 | + | |
342 | + | |
343 | +__kernel_size_t my_strcspn(const char *s1,const char *s2) | |
344 | +{ | |
345 | + char *scan1; | |
346 | + char *scan2; | |
347 | + int count; | |
348 | + | |
349 | + count = 0; | |
350 | + for (scan1 = (char *)s1; *scan1 != '\0'; scan1++) { | |
351 | + for (scan2 = (char *)s2; *scan2 != '\0';) /* ++ moved down. */ | |
352 | + if (*scan1 == *scan2++) | |
353 | + return(count); | |
354 | + count++; | |
355 | + } | |
356 | + return(count); | |
357 | +} | |
358 | + | |
359 | +/* | |
360 | + - regcomp - compile a regular expression into internal code | |
361 | + * | |
362 | + * We can't allocate space until we know how big the compiled form will be, | |
363 | + * but we can't compile it (and thus know how big it is) until we've got a | |
364 | + * place to put the code. So we cheat: we compile it twice, once with code | |
365 | + * generation turned off and size counting turned on, and once "for real". | |
366 | + * This also means that we don't allocate space until we are sure that the | |
367 | + * thing really will compile successfully, and we never have to move the | |
368 | + * code and thus invalidate pointers into it. (Note that it has to be in | |
369 | + * one piece because free() must be able to free it all.) | |
370 | + * | |
371 | + * Beware that the optimization-preparation code in here knows about some | |
372 | + * of the structure of the compiled regexp. | |
373 | + */ | |
374 | +regexp * | |
375 | +regcomp(char *exp,int *patternsize) | |
376 | +{ | |
377 | + register regexp *r; | |
378 | + register char *scan; | |
379 | + register char *longest; | |
380 | + register int len; | |
381 | + int flags; | |
382 | + struct match_globals g; | |
383 | + | |
384 | + /* commented out by ethan | |
385 | + extern char *malloc(); | |
386 | + */ | |
387 | + | |
388 | + if (exp == NULL) | |
389 | + FAIL("NULL argument"); | |
390 | + | |
391 | + /* First pass: determine size, legality. */ | |
392 | + g.regparse = exp; | |
393 | + g.regnpar = 1; | |
394 | + g.regsize = 0L; | |
395 | + g.regcode = &g.regdummy; | |
396 | + regc(&g, MAGIC); | |
397 | + if (reg(&g, 0, &flags) == NULL) | |
398 | + return(NULL); | |
399 | + | |
400 | + /* Small enough for pointer-storage convention? */ | |
401 | + if (g.regsize >= 32767L) /* Probably could be 65535L. */ | |
402 | + FAIL("regexp too big"); | |
403 | + | |
404 | + /* Allocate space. */ | |
405 | + *patternsize=sizeof(regexp) + (unsigned)g.regsize; | |
406 | + r = (regexp *)malloc(sizeof(regexp) + (unsigned)g.regsize); | |
407 | + if (r == NULL) | |
408 | + FAIL("out of space"); | |
409 | + | |
410 | + /* Second pass: emit code. */ | |
411 | + g.regparse = exp; | |
412 | + g.regnpar = 1; | |
413 | + g.regcode = r->program; | |
414 | + regc(&g, MAGIC); | |
415 | + if (reg(&g, 0, &flags) == NULL) | |
416 | + return(NULL); | |
417 | + | |
418 | + /* Dig out information for optimizations. */ | |
419 | + r->regstart = '\0'; /* Worst-case defaults. */ | |
420 | + r->reganch = 0; | |
421 | + r->regmust = NULL; | |
422 | + r->regmlen = 0; | |
423 | + scan = r->program+1; /* First BRANCH. */ | |
424 | + if (OP(regnext(&g, scan)) == END) { /* Only one top-level choice. */ | |
425 | + scan = OPERAND(scan); | |
426 | + | |
427 | + /* Starting-point info. */ | |
428 | + if (OP(scan) == EXACTLY) | |
429 | + r->regstart = *OPERAND(scan); | |
430 | + else if (OP(scan) == BOL) | |
431 | + r->reganch++; | |
432 | + | |
433 | + /* | |
434 | + * If there's something expensive in the r.e., find the | |
435 | + * longest literal string that must appear and make it the | |
436 | + * regmust. Resolve ties in favor of later strings, since | |
437 | + * the regstart check works with the beginning of the r.e. | |
438 | + * and avoiding duplication strengthens checking. Not a | |
439 | + * strong reason, but sufficient in the absence of others. | |
440 | + */ | |
441 | + if (flags&SPSTART) { | |
442 | + longest = NULL; | |
443 | + len = 0; | |
444 | + for (; scan != NULL; scan = regnext(&g, scan)) | |
445 | + if (OP(scan) == EXACTLY && strlen(OPERAND(scan)) >= len) { | |
446 | + longest = OPERAND(scan); | |
447 | + len = strlen(OPERAND(scan)); | |
448 | + } | |
449 | + r->regmust = longest; | |
450 | + r->regmlen = len; | |
451 | + } | |
452 | + } | |
453 | + | |
454 | + return(r); | |
455 | +} | |
456 | + | |
457 | +/* | |
458 | + - reg - regular expression, i.e. main body or parenthesized thing | |
459 | + * | |
460 | + * Caller must absorb opening parenthesis. | |
461 | + * | |
462 | + * Combining parenthesis handling with the base level of regular expression | |
463 | + * is a trifle forced, but the need to tie the tails of the branches to what | |
464 | + * follows makes it hard to avoid. | |
465 | + */ | |
466 | +static char * | |
467 | +reg(struct match_globals *g, int paren, int *flagp /* Parenthesized? */ ) | |
468 | +{ | |
469 | + register char *ret; | |
470 | + register char *br; | |
471 | + register char *ender; | |
472 | + register int parno = 0; /* 0 makes gcc happy */ | |
473 | + int flags; | |
474 | + | |
475 | + *flagp = HASWIDTH; /* Tentatively. */ | |
476 | + | |
477 | + /* Make an OPEN node, if parenthesized. */ | |
478 | + if (paren) { | |
479 | + if (g->regnpar >= NSUBEXP) | |
480 | + FAIL("too many ()"); | |
481 | + parno = g->regnpar; | |
482 | + g->regnpar++; | |
483 | + ret = regnode(g, OPEN+parno); | |
484 | + } else | |
485 | + ret = NULL; | |
486 | + | |
487 | + /* Pick up the branches, linking them together. */ | |
488 | + br = regbranch(g, &flags); | |
489 | + if (br == NULL) | |
490 | + return(NULL); | |
491 | + if (ret != NULL) | |
492 | + regtail(g, ret, br); /* OPEN -> first. */ | |
493 | + else | |
494 | + ret = br; | |
495 | + if (!(flags&HASWIDTH)) | |
496 | + *flagp &= ~HASWIDTH; | |
497 | + *flagp |= flags&SPSTART; | |
498 | + while (*g->regparse == '|') { | |
499 | + g->regparse++; | |
500 | + br = regbranch(g, &flags); | |
501 | + if (br == NULL) | |
502 | + return(NULL); | |
503 | + regtail(g, ret, br); /* BRANCH -> BRANCH. */ | |
504 | + if (!(flags&HASWIDTH)) | |
505 | + *flagp &= ~HASWIDTH; | |
506 | + *flagp |= flags&SPSTART; | |
507 | + } | |
508 | + | |
509 | + /* Make a closing node, and hook it on the end. */ | |
510 | + ender = regnode(g, (paren) ? CLOSE+parno : END); | |
511 | + regtail(g, ret, ender); | |
512 | + | |
513 | + /* Hook the tails of the branches to the closing node. */ | |
514 | + for (br = ret; br != NULL; br = regnext(g, br)) | |
515 | + regoptail(g, br, ender); | |
516 | + | |
517 | + /* Check for proper termination. */ | |
518 | + if (paren && *g->regparse++ != ')') { | |
519 | + FAIL("unmatched ()"); | |
520 | + } else if (!paren && *g->regparse != '\0') { | |
521 | + if (*g->regparse == ')') { | |
522 | + FAIL("unmatched ()"); | |
523 | + } else | |
524 | + FAIL("junk on end"); /* "Can't happen". */ | |
525 | + /* NOTREACHED */ | |
526 | + } | |
527 | + | |
528 | + return(ret); | |
529 | +} | |
530 | + | |
531 | +/* | |
532 | + - regbranch - one alternative of an | operator | |
533 | + * | |
534 | + * Implements the concatenation operator. | |
535 | + */ | |
536 | +static char * | |
537 | +regbranch(struct match_globals *g, int *flagp) | |
538 | +{ | |
539 | + register char *ret; | |
540 | + register char *chain; | |
541 | + register char *latest; | |
542 | + int flags; | |
543 | + | |
544 | + *flagp = WORST; /* Tentatively. */ | |
545 | + | |
546 | + ret = regnode(g, BRANCH); | |
547 | + chain = NULL; | |
548 | + while (*g->regparse != '\0' && *g->regparse != '|' && *g->regparse != ')') { | |
549 | + latest = regpiece(g, &flags); | |
550 | + if (latest == NULL) | |
551 | + return(NULL); | |
552 | + *flagp |= flags&HASWIDTH; | |
553 | + if (chain == NULL) /* First piece. */ | |
554 | + *flagp |= flags&SPSTART; | |
555 | + else | |
556 | + regtail(g, chain, latest); | |
557 | + chain = latest; | |
558 | + } | |
559 | + if (chain == NULL) /* Loop ran zero times. */ | |
560 | + (void) regnode(g, NOTHING); | |
561 | + | |
562 | + return(ret); | |
563 | +} | |
564 | + | |
565 | +/* | |
566 | + - regpiece - something followed by possible [*+?] | |
567 | + * | |
568 | + * Note that the branching code sequences used for ? and the general cases | |
569 | + * of * and + are somewhat optimized: they use the same NOTHING node as | |
570 | + * both the endmarker for their branch list and the body of the last branch. | |
571 | + * It might seem that this node could be dispensed with entirely, but the | |
572 | + * endmarker role is not redundant. | |
573 | + */ | |
574 | +static char * | |
575 | +regpiece(struct match_globals *g, int *flagp) | |
576 | +{ | |
577 | + register char *ret; | |
578 | + register char op; | |
579 | + register char *next; | |
580 | + int flags; | |
581 | + | |
582 | + ret = regatom(g, &flags); | |
583 | + if (ret == NULL) | |
584 | + return(NULL); | |
585 | + | |
586 | + op = *g->regparse; | |
587 | + if (!ISMULT(op)) { | |
588 | + *flagp = flags; | |
589 | + return(ret); | |
590 | + } | |
591 | + | |
592 | + if (!(flags&HASWIDTH) && op != '?') | |
593 | + FAIL("*+ operand could be empty"); | |
594 | + *flagp = (op != '+') ? (WORST|SPSTART) : (WORST|HASWIDTH); | |
595 | + | |
596 | + if (op == '*' && (flags&SIMPLE)) | |
597 | + reginsert(g, STAR, ret); | |
598 | + else if (op == '*') { | |
599 | + /* Emit x* as (x&|), where & means "self". */ | |
600 | + reginsert(g, BRANCH, ret); /* Either x */ | |
601 | + regoptail(g, ret, regnode(g, BACK)); /* and loop */ | |
602 | + regoptail(g, ret, ret); /* back */ | |
603 | + regtail(g, ret, regnode(g, BRANCH)); /* or */ | |
604 | + regtail(g, ret, regnode(g, NOTHING)); /* null. */ | |
605 | + } else if (op == '+' && (flags&SIMPLE)) | |
606 | + reginsert(g, PLUS, ret); | |
607 | + else if (op == '+') { | |
608 | + /* Emit x+ as x(&|), where & means "self". */ | |
609 | + next = regnode(g, BRANCH); /* Either */ | |
610 | + regtail(g, ret, next); | |
611 | + regtail(g, regnode(g, BACK), ret); /* loop back */ | |
612 | + regtail(g, next, regnode(g, BRANCH)); /* or */ | |
613 | + regtail(g, ret, regnode(g, NOTHING)); /* null. */ | |
614 | + } else if (op == '?') { | |
615 | + /* Emit x? as (x|) */ | |
616 | + reginsert(g, BRANCH, ret); /* Either x */ | |
617 | + regtail(g, ret, regnode(g, BRANCH)); /* or */ | |
618 | + next = regnode(g, NOTHING); /* null. */ | |
619 | + regtail(g, ret, next); | |
620 | + regoptail(g, ret, next); | |
621 | + } | |
622 | + g->regparse++; | |
623 | + if (ISMULT(*g->regparse)) | |
624 | + FAIL("nested *?+"); | |
625 | + | |
626 | + return(ret); | |
627 | +} | |
628 | + | |
629 | +/* | |
630 | + - regatom - the lowest level | |
631 | + * | |
632 | + * Optimization: gobbles an entire sequence of ordinary characters so that | |
633 | + * it can turn them into a single node, which is smaller to store and | |
634 | + * faster to run. Backslashed characters are exceptions, each becoming a | |
635 | + * separate node; the code is simpler that way and it's not worth fixing. | |
636 | + */ | |
637 | +static char * | |
638 | +regatom(struct match_globals *g, int *flagp) | |
639 | +{ | |
640 | + register char *ret; | |
641 | + int flags; | |
642 | + | |
643 | + *flagp = WORST; /* Tentatively. */ | |
644 | + | |
645 | + switch (*g->regparse++) { | |
646 | + case '^': | |
647 | + ret = regnode(g, BOL); | |
648 | + break; | |
649 | + case '$': | |
650 | + ret = regnode(g, EOL); | |
651 | + break; | |
652 | + case '.': | |
653 | + ret = regnode(g, ANY); | |
654 | + *flagp |= HASWIDTH|SIMPLE; | |
655 | + break; | |
656 | + case '[': { | |
657 | + register int class; | |
658 | + register int classend; | |
659 | + | |
660 | + if (*g->regparse == '^') { /* Complement of range. */ | |
661 | + ret = regnode(g, ANYBUT); | |
662 | + g->regparse++; | |
663 | + } else | |
664 | + ret = regnode(g, ANYOF); | |
665 | + if (*g->regparse == ']' || *g->regparse == '-') | |
666 | + regc(g, *g->regparse++); | |
667 | + while (*g->regparse != '\0' && *g->regparse != ']') { | |
668 | + if (*g->regparse == '-') { | |
669 | + g->regparse++; | |
670 | + if (*g->regparse == ']' || *g->regparse == '\0') | |
671 | + regc(g, '-'); | |
672 | + else { | |
673 | + class = UCHARAT(g->regparse-2)+1; | |
674 | + classend = UCHARAT(g->regparse); | |
675 | + if (class > classend+1) | |
676 | + FAIL("invalid [] range"); | |
677 | + for (; class <= classend; class++) | |
678 | + regc(g, class); | |
679 | + g->regparse++; | |
680 | + } | |
681 | + } else | |
682 | + regc(g, *g->regparse++); | |
683 | + } | |
684 | + regc(g, '\0'); | |
685 | + if (*g->regparse != ']') | |
686 | + FAIL("unmatched []"); | |
687 | + g->regparse++; | |
688 | + *flagp |= HASWIDTH|SIMPLE; | |
689 | + } | |
690 | + break; | |
691 | + case '(': | |
692 | + ret = reg(g, 1, &flags); | |
693 | + if (ret == NULL) | |
694 | + return(NULL); | |
695 | + *flagp |= flags&(HASWIDTH|SPSTART); | |
696 | + break; | |
697 | + case '\0': | |
698 | + case '|': | |
699 | + case ')': | |
700 | + FAIL("internal urp"); /* Supposed to be caught earlier. */ | |
701 | + break; | |
702 | + case '?': | |
703 | + case '+': | |
704 | + case '*': | |
705 | + FAIL("?+* follows nothing"); | |
706 | + break; | |
707 | + case '\\': | |
708 | + if (*g->regparse == '\0') | |
709 | + FAIL("trailing \\"); | |
710 | + ret = regnode(g, EXACTLY); | |
711 | + regc(g, *g->regparse++); | |
712 | + regc(g, '\0'); | |
713 | + *flagp |= HASWIDTH|SIMPLE; | |
714 | + break; | |
715 | + default: { | |
716 | + register int len; | |
717 | + register char ender; | |
718 | + | |
719 | + g->regparse--; | |
720 | + len = my_strcspn((const char *)g->regparse, (const char *)META); | |
721 | + if (len <= 0) | |
722 | + FAIL("internal disaster"); | |
723 | + ender = *(g->regparse+len); | |
724 | + if (len > 1 && ISMULT(ender)) | |
725 | + len--; /* Back off clear of ?+* operand. */ | |
726 | + *flagp |= HASWIDTH; | |
727 | + if (len == 1) | |
728 | + *flagp |= SIMPLE; | |
729 | + ret = regnode(g, EXACTLY); | |
730 | + while (len > 0) { | |
731 | + regc(g, *g->regparse++); | |
732 | + len--; | |
733 | + } | |
734 | + regc(g, '\0'); | |
735 | + } | |
736 | + break; | |
737 | + } | |
738 | + | |
739 | + return(ret); | |
740 | +} | |
741 | + | |
742 | +/* | |
743 | + - regnode - emit a node | |
744 | + */ | |
745 | +static char * /* Location. */ | |
746 | +regnode(struct match_globals *g, char op) | |
747 | +{ | |
748 | + register char *ret; | |
749 | + register char *ptr; | |
750 | + | |
751 | + ret = g->regcode; | |
752 | + if (ret == &g->regdummy) { | |
753 | + g->regsize += 3; | |
754 | + return(ret); | |
755 | + } | |
756 | + | |
757 | + ptr = ret; | |
758 | + *ptr++ = op; | |
759 | + *ptr++ = '\0'; /* Null "next" pointer. */ | |
760 | + *ptr++ = '\0'; | |
761 | + g->regcode = ptr; | |
762 | + | |
763 | + return(ret); | |
764 | +} | |
765 | + | |
766 | +/* | |
767 | + - regc - emit (if appropriate) a byte of code | |
768 | + */ | |
769 | +static void | |
770 | +regc(struct match_globals *g, char b) | |
771 | +{ | |
772 | + if (g->regcode != &g->regdummy) | |
773 | + *g->regcode++ = b; | |
774 | + else | |
775 | + g->regsize++; | |
776 | +} | |
777 | + | |
778 | +/* | |
779 | + - reginsert - insert an operator in front of already-emitted operand | |
780 | + * | |
781 | + * Means relocating the operand. | |
782 | + */ | |
783 | +static void | |
784 | +reginsert(struct match_globals *g, char op, char* opnd) | |
785 | +{ | |
786 | + register char *src; | |
787 | + register char *dst; | |
788 | + register char *place; | |
789 | + | |
790 | + if (g->regcode == &g->regdummy) { | |
791 | + g->regsize += 3; | |
792 | + return; | |
793 | + } | |
794 | + | |
795 | + src = g->regcode; | |
796 | + g->regcode += 3; | |
797 | + dst = g->regcode; | |
798 | + while (src > opnd) | |
799 | + *--dst = *--src; | |
800 | + | |
801 | + place = opnd; /* Op node, where operand used to be. */ | |
802 | + *place++ = op; | |
803 | + *place++ = '\0'; | |
804 | + *place++ = '\0'; | |
805 | +} | |
806 | + | |
807 | +/* | |
808 | + - regtail - set the next-pointer at the end of a node chain | |
809 | + */ | |
810 | +static void | |
811 | +regtail(struct match_globals *g, char *p, char *val) | |
812 | +{ | |
813 | + register char *scan; | |
814 | + register char *temp; | |
815 | + register int offset; | |
816 | + | |
817 | + if (p == &g->regdummy) | |
818 | + return; | |
819 | + | |
820 | + /* Find last node. */ | |
821 | + scan = p; | |
822 | + for (;;) { | |
823 | + temp = regnext(g, scan); | |
824 | + if (temp == NULL) | |
825 | + break; | |
826 | + scan = temp; | |
827 | + } | |
828 | + | |
829 | + if (OP(scan) == BACK) | |
830 | + offset = scan - val; | |
831 | + else | |
832 | + offset = val - scan; | |
833 | + *(scan+1) = (offset>>8)&0377; | |
834 | + *(scan+2) = offset&0377; | |
835 | +} | |
836 | + | |
837 | +/* | |
838 | + - regoptail - regtail on operand of first argument; nop if operandless | |
839 | + */ | |
840 | +static void | |
841 | +regoptail(struct match_globals *g, char *p, char *val) | |
842 | +{ | |
843 | + /* "Operandless" and "op != BRANCH" are synonymous in practice. */ | |
844 | + if (p == NULL || p == &g->regdummy || OP(p) != BRANCH) | |
845 | + return; | |
846 | + regtail(g, OPERAND(p), val); | |
847 | +} | |
848 | + | |
849 | +/* | |
850 | + * regexec and friends | |
851 | + */ | |
852 | + | |
853 | + | |
854 | +/* | |
855 | + * Forwards. | |
856 | + */ | |
857 | +STATIC int regtry(struct match_globals *g, regexp *prog, char *string); | |
858 | +STATIC int regmatch(struct match_globals *g, char *prog); | |
859 | +STATIC int regrepeat(struct match_globals *g, char *p); | |
860 | + | |
861 | +#ifdef DEBUG | |
862 | +int regnarrate = 0; | |
863 | +void regdump(); | |
864 | +STATIC char *regprop(char *op); | |
865 | +#endif | |
866 | + | |
867 | +/* | |
868 | + - regexec - match a regexp against a string | |
869 | + */ | |
870 | +int | |
871 | +regexec(regexp *prog, char *string) | |
872 | +{ | |
873 | + register char *s; | |
874 | + struct match_globals g; | |
875 | + | |
876 | + /* Be paranoid... */ | |
877 | + if (prog == NULL || string == NULL) { | |
878 | + printk("<3>Regexp: NULL parameter\n"); | |
879 | + return(0); | |
880 | + } | |
881 | + | |
882 | + /* Check validity of program. */ | |
883 | + if (UCHARAT(prog->program) != MAGIC) { | |
884 | + printk("<3>Regexp: corrupted program\n"); | |
885 | + return(0); | |
886 | + } | |
887 | + | |
888 | + /* If there is a "must appear" string, look for it. */ | |
889 | + if (prog->regmust != NULL) { | |
890 | + s = string; | |
891 | + while ((s = strchr(s, prog->regmust[0])) != NULL) { | |
892 | + if (strncmp(s, prog->regmust, prog->regmlen) == 0) | |
893 | + break; /* Found it. */ | |
894 | + s++; | |
895 | + } | |
896 | + if (s == NULL) /* Not present. */ | |
897 | + return(0); | |
898 | + } | |
899 | + | |
900 | + /* Mark beginning of line for ^ . */ | |
901 | + g.regbol = string; | |
902 | + | |
903 | + /* Simplest case: anchored match need be tried only once. */ | |
904 | + if (prog->reganch) | |
905 | + return(regtry(&g, prog, string)); | |
906 | + | |
907 | + /* Messy cases: unanchored match. */ | |
908 | + s = string; | |
909 | + if (prog->regstart != '\0') | |
910 | + /* We know what char it must start with. */ | |
911 | + while ((s = strchr(s, prog->regstart)) != NULL) { | |
912 | + if (regtry(&g, prog, s)) | |
913 | + return(1); | |
914 | + s++; | |
915 | + } | |
916 | + else | |
917 | + /* We don't -- general case. */ | |
918 | + do { | |
919 | + if (regtry(&g, prog, s)) | |
920 | + return(1); | |
921 | + } while (*s++ != '\0'); | |
922 | + | |
923 | + /* Failure. */ | |
924 | + return(0); | |
925 | +} | |
926 | + | |
927 | +/* | |
928 | + - regtry - try match at specific point | |
929 | + */ | |
930 | +static int /* 0 failure, 1 success */ | |
931 | +regtry(struct match_globals *g, regexp *prog, char *string) | |
932 | +{ | |
933 | + register int i; | |
934 | + register char **sp; | |
935 | + register char **ep; | |
936 | + | |
937 | + g->reginput = string; | |
938 | + g->regstartp = prog->startp; | |
939 | + g->regendp = prog->endp; | |
940 | + | |
941 | + sp = prog->startp; | |
942 | + ep = prog->endp; | |
943 | + for (i = NSUBEXP; i > 0; i--) { | |
944 | + *sp++ = NULL; | |
945 | + *ep++ = NULL; | |
946 | + } | |
947 | + if (regmatch(g, prog->program + 1)) { | |
948 | + prog->startp[0] = string; | |
949 | + prog->endp[0] = g->reginput; | |
950 | + return(1); | |
951 | + } else | |
952 | + return(0); | |
953 | +} | |
954 | + | |
955 | +/* | |
956 | + - regmatch - main matching routine | |
957 | + * | |
958 | + * Conceptually the strategy is simple: check to see whether the current | |
959 | + * node matches, call self recursively to see whether the rest matches, | |
960 | + * and then act accordingly. In practice we make some effort to avoid | |
961 | + * recursion, in particular by going through "ordinary" nodes (that don't | |
962 | + * need to know whether the rest of the match failed) by a loop instead of | |
963 | + * by recursion. | |
964 | + */ | |
965 | +static int /* 0 failure, 1 success */ | |
966 | +regmatch(struct match_globals *g, char *prog) | |
967 | +{ | |
968 | + register char *scan = prog; /* Current node. */ | |
969 | + char *next; /* Next node. */ | |
970 | + | |
971 | +#ifdef DEBUG | |
972 | + if (scan != NULL && regnarrate) | |
973 | + fprintf(stderr, "%s(\n", regprop(scan)); | |
974 | +#endif | |
975 | + while (scan != NULL) { | |
976 | +#ifdef DEBUG | |
977 | + if (regnarrate) | |
978 | + fprintf(stderr, "%s...\n", regprop(scan)); | |
979 | +#endif | |
980 | + next = regnext(g, scan); | |
981 | + | |
982 | + switch (OP(scan)) { | |
983 | + case BOL: | |
984 | + if (g->reginput != g->regbol) | |
985 | + return(0); | |
986 | + break; | |
987 | + case EOL: | |
988 | + if (*g->reginput != '\0') | |
989 | + return(0); | |
990 | + break; | |
991 | + case ANY: | |
992 | + if (*g->reginput == '\0') | |
993 | + return(0); | |
994 | + g->reginput++; | |
995 | + break; | |
996 | + case EXACTLY: { | |
997 | + register int len; | |
998 | + register char *opnd; | |
999 | + | |
1000 | + opnd = OPERAND(scan); | |
1001 | + /* Inline the first character, for speed. */ | |
1002 | + if (*opnd != *g->reginput) | |
1003 | + return(0); | |
1004 | + len = strlen(opnd); | |
1005 | + if (len > 1 && strncmp(opnd, g->reginput, len) != 0) | |
1006 | + return(0); | |
1007 | + g->reginput += len; | |
1008 | + } | |
1009 | + break; | |
1010 | + case ANYOF: | |
1011 | + if (*g->reginput == '\0' || strchr(OPERAND(scan), *g->reginput) == NULL) | |
1012 | + return(0); | |
1013 | + g->reginput++; | |
1014 | + break; | |
1015 | + case ANYBUT: | |
1016 | + if (*g->reginput == '\0' || strchr(OPERAND(scan), *g->reginput) != NULL) | |
1017 | + return(0); | |
1018 | + g->reginput++; | |
1019 | + break; | |
1020 | + case NOTHING: | |
1021 | + case BACK: | |
1022 | + break; | |
1023 | + case OPEN+1: | |
1024 | + case OPEN+2: | |
1025 | + case OPEN+3: | |
1026 | + case OPEN+4: | |
1027 | + case OPEN+5: | |
1028 | + case OPEN+6: | |
1029 | + case OPEN+7: | |
1030 | + case OPEN+8: | |
1031 | + case OPEN+9: { | |
1032 | + register int no; | |
1033 | + register char *save; | |
1034 | + | |
1035 | + no = OP(scan) - OPEN; | |
1036 | + save = g->reginput; | |
1037 | + | |
1038 | + if (regmatch(g, next)) { | |
1039 | + /* | |
1040 | + * Don't set startp if some later | |
1041 | + * invocation of the same parentheses | |
1042 | + * already has. | |
1043 | + */ | |
1044 | + if (g->regstartp[no] == NULL) | |
1045 | + g->regstartp[no] = save; | |
1046 | + return(1); | |
1047 | + } else | |
1048 | + return(0); | |
1049 | + } | |
1050 | + break; | |
1051 | + case CLOSE+1: | |
1052 | + case CLOSE+2: | |
1053 | + case CLOSE+3: | |
1054 | + case CLOSE+4: | |
1055 | + case CLOSE+5: | |
1056 | + case CLOSE+6: | |
1057 | + case CLOSE+7: | |
1058 | + case CLOSE+8: | |
1059 | + case CLOSE+9: | |
1060 | + { | |
1061 | + register int no; | |
1062 | + register char *save; | |
1063 | + | |
1064 | + no = OP(scan) - CLOSE; | |
1065 | + save = g->reginput; | |
1066 | + | |
1067 | + if (regmatch(g, next)) { | |
1068 | + /* | |
1069 | + * Don't set endp if some later | |
1070 | + * invocation of the same parentheses | |
1071 | + * already has. | |
1072 | + */ | |
1073 | + if (g->regendp[no] == NULL) | |
1074 | + g->regendp[no] = save; | |
1075 | + return(1); | |
1076 | + } else | |
1077 | + return(0); | |
1078 | + } | |
1079 | + break; | |
1080 | + case BRANCH: { | |
1081 | + register char *save; | |
1082 | + | |
1083 | + if (OP(next) != BRANCH) /* No choice. */ | |
1084 | + next = OPERAND(scan); /* Avoid recursion. */ | |
1085 | + else { | |
1086 | + do { | |
1087 | + save = g->reginput; | |
1088 | + if (regmatch(g, OPERAND(scan))) | |
1089 | + return(1); | |
1090 | + g->reginput = save; | |
1091 | + scan = regnext(g, scan); | |
1092 | + } while (scan != NULL && OP(scan) == BRANCH); | |
1093 | + return(0); | |
1094 | + /* NOTREACHED */ | |
1095 | + } | |
1096 | + } | |
1097 | + break; | |
1098 | + case STAR: | |
1099 | + case PLUS: { | |
1100 | + register char nextch; | |
1101 | + register int no; | |
1102 | + register char *save; | |
1103 | + register int min; | |
1104 | + | |
1105 | + /* | |
1106 | + * Lookahead to avoid useless match attempts | |
1107 | + * when we know what character comes next. | |
1108 | + */ | |
1109 | + nextch = '\0'; | |
1110 | + if (OP(next) == EXACTLY) | |
1111 | + nextch = *OPERAND(next); | |
1112 | + min = (OP(scan) == STAR) ? 0 : 1; | |
1113 | + save = g->reginput; | |
1114 | + no = regrepeat(g, OPERAND(scan)); | |
1115 | + while (no >= min) { | |
1116 | + /* If it could work, try it. */ | |
1117 | + if (nextch == '\0' || *g->reginput == nextch) | |
1118 | + if (regmatch(g, next)) | |
1119 | + return(1); | |
1120 | + /* Couldn't or didn't -- back up. */ | |
1121 | + no--; | |
1122 | + g->reginput = save + no; | |
1123 | + } | |
1124 | + return(0); | |
1125 | + } | |
1126 | + break; | |
1127 | + case END: | |
1128 | + return(1); /* Success! */ | |
1129 | + break; | |
1130 | + default: | |
1131 | + printk("<3>Regexp: memory corruption\n"); | |
1132 | + return(0); | |
1133 | + break; | |
1134 | + } | |
1135 | + | |
1136 | + scan = next; | |
1137 | + } | |
1138 | + | |
1139 | + /* | |
1140 | + * We get here only if there's trouble -- normally "case END" is | |
1141 | + * the terminating point. | |
1142 | + */ | |
1143 | + printk("<3>Regexp: corrupted pointers\n"); | |
1144 | + return(0); | |
1145 | +} | |
1146 | + | |
1147 | +/* | |
1148 | + - regrepeat - repeatedly match something simple, report how many | |
1149 | + */ | |
1150 | +static int | |
1151 | +regrepeat(struct match_globals *g, char *p) | |
1152 | +{ | |
1153 | + register int count = 0; | |
1154 | + register char *scan; | |
1155 | + register char *opnd; | |
1156 | + | |
1157 | + scan = g->reginput; | |
1158 | + opnd = OPERAND(p); | |
1159 | + switch (OP(p)) { | |
1160 | + case ANY: | |
1161 | + count = strlen(scan); | |
1162 | + scan += count; | |
1163 | + break; | |
1164 | + case EXACTLY: | |
1165 | + while (*opnd == *scan) { | |
1166 | + count++; | |
1167 | + scan++; | |
1168 | + } | |
1169 | + break; | |
1170 | + case ANYOF: | |
1171 | + while (*scan != '\0' && strchr(opnd, *scan) != NULL) { | |
1172 | + count++; | |
1173 | + scan++; | |
1174 | + } | |
1175 | + break; | |
1176 | + case ANYBUT: | |
1177 | + while (*scan != '\0' && strchr(opnd, *scan) == NULL) { | |
1178 | + count++; | |
1179 | + scan++; | |
1180 | + } | |
1181 | + break; | |
1182 | + default: /* Oh dear. Called inappropriately. */ | |
1183 | + printk("<3>Regexp: internal foulup\n"); | |
1184 | + count = 0; /* Best compromise. */ | |
1185 | + break; | |
1186 | + } | |
1187 | + g->reginput = scan; | |
1188 | + | |
1189 | + return(count); | |
1190 | +} | |
1191 | + | |
1192 | +/* | |
1193 | + - regnext - dig the "next" pointer out of a node | |
1194 | + */ | |
1195 | +static char* | |
1196 | +regnext(struct match_globals *g, char *p) | |
1197 | +{ | |
1198 | + register int offset; | |
1199 | + | |
1200 | + if (p == &g->regdummy) | |
1201 | + return(NULL); | |
1202 | + | |
1203 | + offset = NEXT(p); | |
1204 | + if (offset == 0) | |
1205 | + return(NULL); | |
1206 | + | |
1207 | + if (OP(p) == BACK) | |
1208 | + return(p-offset); | |
1209 | + else | |
1210 | + return(p+offset); | |
1211 | +} | |
1212 | + | |
1213 | +#ifdef DEBUG | |
1214 | + | |
1215 | +STATIC char *regprop(); | |
1216 | + | |
1217 | +/* | |
1218 | + - regdump - dump a regexp onto stdout in vaguely comprehensible form | |
1219 | + */ | |
1220 | +void | |
1221 | +regdump(regexp *r) | |
1222 | +{ | |
1223 | + register char *s; | |
1224 | + register char op = EXACTLY; /* Arbitrary non-END op. */ | |
1225 | + register char *next; | |
1226 | + /* extern char *strchr(); */ | |
1227 | + | |
1228 | + | |
1229 | + s = r->program + 1; | |
1230 | + while (op != END) { /* While that wasn't END last time... */ | |
1231 | + op = OP(s); | |
1232 | + printf("%2d%s", s-r->program, regprop(s)); /* Where, what. */ | |
1233 | + next = regnext(s); | |
1234 | + if (next == NULL) /* Next ptr. */ | |
1235 | + printf("(0)"); | |
1236 | + else | |
1237 | + printf("(%d)", (s-r->program)+(next-s)); | |
1238 | + s += 3; | |
1239 | + if (op == ANYOF || op == ANYBUT || op == EXACTLY) { | |
1240 | + /* Literal string, where present. */ | |
1241 | + while (*s != '\0') { | |
1242 | + putchar(*s); | |
1243 | + s++; | |
1244 | + } | |
1245 | + s++; | |
1246 | + } | |
1247 | + putchar('\n'); | |
1248 | + } | |
1249 | + | |
1250 | + /* Header fields of interest. */ | |
1251 | + if (r->regstart != '\0') | |
1252 | + printf("start `%c' ", r->regstart); | |
1253 | + if (r->reganch) | |
1254 | + printf("anchored "); | |
1255 | + if (r->regmust != NULL) | |
1256 | + printf("must have \"%s\"", r->regmust); | |
1257 | + printf("\n"); | |
1258 | +} | |
1259 | + | |
1260 | +/* | |
1261 | + - regprop - printable representation of opcode | |
1262 | + */ | |
1263 | +static char * | |
1264 | +regprop(char *op) | |
1265 | +{ | |
1266 | +#define BUFLEN 50 | |
1267 | + register char *p; | |
1268 | + static char buf[BUFLEN]; | |
1269 | + | |
1270 | + strcpy(buf, ":"); | |
1271 | + | |
1272 | + switch (OP(op)) { | |
1273 | + case BOL: | |
1274 | + p = "BOL"; | |
1275 | + break; | |
1276 | + case EOL: | |
1277 | + p = "EOL"; | |
1278 | + break; | |
1279 | + case ANY: | |
1280 | + p = "ANY"; | |
1281 | + break; | |
1282 | + case ANYOF: | |
1283 | + p = "ANYOF"; | |
1284 | + break; | |
1285 | + case ANYBUT: | |
1286 | + p = "ANYBUT"; | |
1287 | + break; | |
1288 | + case BRANCH: | |
1289 | + p = "BRANCH"; | |
1290 | + break; | |
1291 | + case EXACTLY: | |
1292 | + p = "EXACTLY"; | |
1293 | + break; | |
1294 | + case NOTHING: | |
1295 | + p = "NOTHING"; | |
1296 | + break; | |
1297 | + case BACK: | |
1298 | + p = "BACK"; | |
1299 | + break; | |
1300 | + case END: | |
1301 | + p = "END"; | |
1302 | + break; | |
1303 | + case OPEN+1: | |
1304 | + case OPEN+2: | |
1305 | + case OPEN+3: | |
1306 | + case OPEN+4: | |
1307 | + case OPEN+5: | |
1308 | + case OPEN+6: | |
1309 | + case OPEN+7: | |
1310 | + case OPEN+8: | |
1311 | + case OPEN+9: | |
1312 | + snprintf(buf+strlen(buf),BUFLEN-strlen(buf), "OPEN%d", OP(op)-OPEN); | |
1313 | + p = NULL; | |
1314 | + break; | |
1315 | + case CLOSE+1: | |
1316 | + case CLOSE+2: | |
1317 | + case CLOSE+3: | |
1318 | + case CLOSE+4: | |
1319 | + case CLOSE+5: | |
1320 | + case CLOSE+6: | |
1321 | + case CLOSE+7: | |
1322 | + case CLOSE+8: | |
1323 | + case CLOSE+9: | |
1324 | + snprintf(buf+strlen(buf),BUFLEN-strlen(buf), "CLOSE%d", OP(op)-CLOSE); | |
1325 | + p = NULL; | |
1326 | + break; | |
1327 | + case STAR: | |
1328 | + p = "STAR"; | |
1329 | + break; | |
1330 | + case PLUS: | |
1331 | + p = "PLUS"; | |
1332 | + break; | |
1333 | + default: | |
1334 | + printk("<3>Regexp: corrupted opcode\n"); | |
1335 | + break; | |
1336 | + } | |
1337 | + if (p != NULL) | |
1338 | + strncat(buf, p, BUFLEN-strlen(buf)); | |
1339 | + return(buf); | |
1340 | +} | |
1341 | +#endif | |
1342 | + | |
1343 | + | |
1344 | diff -Naur linux-3.14.22.org/net/netfilter/regexp/regexp.h linux-3.14.22/net/netfilter/regexp/regexp.h | |
1345 | --- linux-3.14.22.org/net/netfilter/regexp/regexp.h 1970-01-01 01:00:00.000000000 +0100 | |
1346 | +++ linux-3.14.22/net/netfilter/regexp/regexp.h 2014-10-24 15:33:21.285274399 +0200 | |
1347 | @@ -0,0 +1,41 @@ | |
1348 | +/* | |
1349 | + * Definitions etc. for regexp(3) routines. | |
1350 | + * | |
1351 | + * Caveat: this is V8 regexp(3) [actually, a reimplementation thereof], | |
1352 | + * not the System V one. | |
1353 | + */ | |
1354 | + | |
1355 | +#ifndef REGEXP_H | |
1356 | +#define REGEXP_H | |
1357 | + | |
1358 | + | |
1359 | +/* | |
1360 | +http://www.opensource.apple.com/darwinsource/10.3/expect-1/expect/expect.h , | |
1361 | +which contains a version of this library, says: | |
1362 | + | |
1363 | + * | |
1364 | + * NSUBEXP must be at least 10, and no greater than 117 or the parser | |
1365 | + * will not work properly. | |
1366 | + * | |
1367 | + | |
1368 | +However, it looks rather like this library is limited to 10. If you think | |
1369 | +otherwise, let us know. | |
1370 | +*/ | |
1371 | + | |
1372 | +#define NSUBEXP 10 | |
1373 | +typedef struct regexp { | |
1374 | + char *startp[NSUBEXP]; | |
1375 | + char *endp[NSUBEXP]; | |
1376 | + char regstart; /* Internal use only. */ | |
1377 | + char reganch; /* Internal use only. */ | |
1378 | + char *regmust; /* Internal use only. */ | |
1379 | + int regmlen; /* Internal use only. */ | |
1380 | + char program[1]; /* Unwarranted chumminess with compiler. */ | |
1381 | +} regexp; | |
1382 | + | |
1383 | +regexp * regcomp(char *exp, int *patternsize); | |
1384 | +int regexec(regexp *prog, char *string); | |
1385 | +void regsub(regexp *prog, char *source, char *dest); | |
1386 | +void regerror(char *s); | |
1387 | + | |
1388 | +#endif | |
1389 | diff -Naur linux-3.14.22.org/net/netfilter/regexp/regmagic.h linux-3.14.22/net/netfilter/regexp/regmagic.h | |
1390 | --- linux-3.14.22.org/net/netfilter/regexp/regmagic.h 1970-01-01 01:00:00.000000000 +0100 | |
1391 | +++ linux-3.14.22/net/netfilter/regexp/regmagic.h 2014-10-24 15:33:21.285274399 +0200 | |
1392 | @@ -0,0 +1,5 @@ | |
1393 | +/* | |
1394 | + * The first byte of the regexp internal "program" is actually this magic | |
1395 | + * number; the start node begins in the second byte. | |
1396 | + */ | |
1397 | +#define MAGIC 0234 | |
1398 | diff -Naur linux-3.14.22.org/net/netfilter/regexp/regsub.c linux-3.14.22/net/netfilter/regexp/regsub.c | |
1399 | --- linux-3.14.22.org/net/netfilter/regexp/regsub.c 1970-01-01 01:00:00.000000000 +0100 | |
1400 | +++ linux-3.14.22/net/netfilter/regexp/regsub.c 2014-10-24 15:33:21.285274399 +0200 | |
1401 | @@ -0,0 +1,95 @@ | |
1402 | +/* | |
1403 | + * regsub | |
1404 | + * @(#)regsub.c 1.3 of 2 April 86 | |
1405 | + * | |
1406 | + * Copyright (c) 1986 by University of Toronto. | |
1407 | + * Written by Henry Spencer. Not derived from licensed software. | |
1408 | + * | |
1409 | + * Permission is granted to anyone to use this software for any | |
1410 | + * purpose on any computer system, and to redistribute it freely, | |
1411 | + * subject to the following restrictions: | |
1412 | + * | |
1413 | + * 1. The author is not responsible for the consequences of use of | |
1414 | + * this software, no matter how awful, even if they arise | |
1415 | + * from defects in it. | |
1416 | + * | |
1417 | + * 2. The origin of this software must not be misrepresented, either | |
1418 | + * by explicit claim or by omission. | |
1419 | + * | |
1420 | + * 3. Altered versions must be plainly marked as such, and must not | |
1421 | + * be misrepresented as being the original software. | |
1422 | + * | |
1423 | + * | |
1424 | + * This code was modified by Ethan Sommer to work within the kernel | |
1425 | + * (it now uses kmalloc etc..) | |
1426 | + * | |
1427 | + */ | |
1428 | +#include "regexp.h" | |
1429 | +#include "regmagic.h" | |
1430 | +#include <linux/string.h> | |
1431 | + | |
1432 | + | |
1433 | +#ifndef CHARBITS | |
1434 | +#define UCHARAT(p) ((int)*(unsigned char *)(p)) | |
1435 | +#else | |
1436 | +#define UCHARAT(p) ((int)*(p)&CHARBITS) | |
1437 | +#endif | |
1438 | + | |
1439 | +#if 0 | |
1440 | +//void regerror(char * s) | |
1441 | +//{ | |
1442 | +// printk("regexp(3): %s", s); | |
1443 | +// /* NOTREACHED */ | |
1444 | +//} | |
1445 | +#endif | |
1446 | + | |
1447 | +/* | |
1448 | + - regsub - perform substitutions after a regexp match | |
1449 | + */ | |
1450 | +void | |
1451 | +regsub(regexp * prog, char * source, char * dest) | |
1452 | +{ | |
1453 | + register char *src; | |
1454 | + register char *dst; | |
1455 | + register char c; | |
1456 | + register int no; | |
1457 | + register int len; | |
1458 | + | |
1459 | + /* Not necessary and gcc doesn't like it -MLS */ | |
1460 | + /*extern char *strncpy();*/ | |
1461 | + | |
1462 | + if (prog == NULL || source == NULL || dest == NULL) { | |
1463 | + regerror("NULL parm to regsub"); | |
1464 | + return; | |
1465 | + } | |
1466 | + if (UCHARAT(prog->program) != MAGIC) { | |
1467 | + regerror("damaged regexp fed to regsub"); | |
1468 | + return; | |
1469 | + } | |
1470 | + | |
1471 | + src = source; | |
1472 | + dst = dest; | |
1473 | + while ((c = *src++) != '\0') { | |
1474 | + if (c == '&') | |
1475 | + no = 0; | |
1476 | + else if (c == '\\' && '0' <= *src && *src <= '9') | |
1477 | + no = *src++ - '0'; | |
1478 | + else | |
1479 | + no = -1; | |
1480 | + | |
1481 | + if (no < 0) { /* Ordinary character. */ | |
1482 | + if (c == '\\' && (*src == '\\' || *src == '&')) | |
1483 | + c = *src++; | |
1484 | + *dst++ = c; | |
1485 | + } else if (prog->startp[no] != NULL && prog->endp[no] != NULL) { | |
1486 | + len = prog->endp[no] - prog->startp[no]; | |
1487 | + (void) strncpy(dst, prog->startp[no], len); | |
1488 | + dst += len; | |
1489 | + if (len != 0 && *(dst-1) == '\0') { /* strncpy hit NUL. */ | |
1490 | + regerror("damaged match string"); | |
1491 | + return; | |
1492 | + } | |
1493 | + } | |
1494 | + } | |
1495 | + *dst++ = '\0'; | |
1496 | +} | |
1497 | diff -Naur linux-3.14.22.org/net/netfilter/xt_layer7.c linux-3.14.22/net/netfilter/xt_layer7.c | |
1498 | --- linux-3.14.22.org/net/netfilter/xt_layer7.c 1970-01-01 01:00:00.000000000 +0100 | |
1499 | +++ linux-3.14.22/net/netfilter/xt_layer7.c 2014-10-24 15:33:21.285274399 +0200 | |
1500 | @@ -0,0 +1,665 @@ | |
1501 | +/* | |
1502 | + Kernel module to match application layer (OSI layer 7) data in connections. | |
1503 | + | |
1504 | + http://l7-filter.sf.net | |
1505 | + | |
1506 | + (C) 2003-2009 Matthew Strait and Ethan Sommer. | |
1507 | + | |
1508 | + This program is free software; you can redistribute it and/or | |
1509 | + modify it under the terms of the GNU General Public License | |
1510 | + as published by the Free Software Foundation; either version | |
1511 | + 2 of the License, or (at your option) any later version. | |
1512 | + http://www.gnu.org/licenses/gpl.txt | |
1513 | + | |
1514 | + Based on ipt_string.c (C) 2000 Emmanuel Roger <winfield@freegates.be>, | |
1515 | + xt_helper.c (C) 2002 Harald Welte and cls_layer7.c (C) 2003 Matthew Strait, | |
1516 | + Ethan Sommer, Justin Levandoski. | |
1517 | +*/ | |
1518 | + | |
1519 | +#include <linux/spinlock.h> | |
1520 | +#include <linux/version.h> | |
1521 | +#include <net/ip.h> | |
1522 | +#include <net/tcp.h> | |
1523 | +#include <linux/module.h> | |
1524 | +#include <linux/seq_file.h> | |
1525 | +#include <linux/skbuff.h> | |
1526 | +#include <linux/netfilter.h> | |
1527 | +#include <net/netfilter/nf_conntrack.h> | |
1528 | +#include <net/netfilter/nf_conntrack_core.h> | |
1529 | +#include <net/netfilter/nf_conntrack_extend.h> | |
1530 | +#include <net/netfilter/nf_conntrack_acct.h> | |
1531 | +#include <linux/netfilter/x_tables.h> | |
1532 | +#include <linux/netfilter/xt_layer7.h> | |
1533 | +#include <linux/ctype.h> | |
1534 | +#include <linux/proc_fs.h> | |
1535 | + | |
1536 | +#include "regexp/regexp.c" | |
1537 | + | |
1538 | +MODULE_LICENSE("GPL"); | |
1539 | +MODULE_AUTHOR("Matthew Strait <quadong@users.sf.net>, Ethan Sommer <sommere@users.sf.net>"); | |
1540 | +MODULE_DESCRIPTION("iptables application layer match module"); | |
1541 | +MODULE_ALIAS("ipt_layer7"); | |
1542 | +MODULE_VERSION("2.21"); | |
1543 | + | |
1544 | +static int maxdatalen = 2048; // this is the default | |
1545 | +module_param(maxdatalen, int, 0444); | |
1546 | +MODULE_PARM_DESC(maxdatalen, "maximum bytes of data looked at by l7-filter"); | |
1547 | +#ifdef CONFIG_NETFILTER_XT_MATCH_LAYER7_DEBUG | |
1548 | + #define DPRINTK(format,args...) printk(format,##args) | |
1549 | +#else | |
1550 | + #define DPRINTK(format,args...) | |
1551 | +#endif | |
1552 | + | |
1553 | +/* Number of packets whose data we look at. | |
1554 | +This can be modified through /proc/net/layer7_numpackets */ | |
1555 | +static int num_packets = 10; | |
1556 | + | |
1557 | +static struct pattern_cache { | |
1558 | + char * regex_string; | |
1559 | + regexp * pattern; | |
1560 | + struct pattern_cache * next; | |
1561 | +} * first_pattern_cache = NULL; | |
1562 | + | |
1563 | +DEFINE_SPINLOCK(l7_lock); | |
1564 | + | |
1565 | +static int total_acct_packets(struct nf_conn *ct) | |
1566 | +{ | |
1567 | + struct nf_conn_counter *acct; | |
1568 | + | |
1569 | + BUG_ON(ct == NULL); | |
1570 | + acct = nf_conn_acct_find(ct); | |
1571 | + if (!acct) | |
1572 | + return 0; | |
1573 | + return (atomic64_read(&acct[IP_CT_DIR_ORIGINAL].packets) + atomic64_read(&acct[IP_CT_DIR_REPLY].packets)); | |
1574 | +} | |
1575 | + | |
1576 | +#ifdef CONFIG_IP_NF_MATCH_LAYER7_DEBUG | |
1577 | +/* Converts an unfriendly string into a friendly one by | |
1578 | +replacing unprintables with periods and all whitespace with " ". */ | |
1579 | +static char * friendly_print(unsigned char * s) | |
1580 | +{ | |
1581 | + char * f = kmalloc(strlen(s) + 1, GFP_ATOMIC); | |
1582 | + int i; | |
1583 | + | |
1584 | + if(!f) { | |
1585 | + if (net_ratelimit()) | |
1586 | + printk(KERN_ERR "layer7: out of memory in " | |
1587 | + "friendly_print, bailing.\n"); | |
1588 | + return NULL; | |
1589 | + } | |
1590 | + | |
1591 | + for(i = 0; i < strlen(s); i++){ | |
1592 | + if(isprint(s[i]) && s[i] < 128) f[i] = s[i]; | |
1593 | + else if(isspace(s[i])) f[i] = ' '; | |
1594 | + else f[i] = '.'; | |
1595 | + } | |
1596 | + f[i] = '\0'; | |
1597 | + return f; | |
1598 | +} | |
1599 | + | |
1600 | +static char dec2hex(int i) | |
1601 | +{ | |
1602 | + switch (i) { | |
1603 | + case 0 ... 9: | |
1604 | + return (i + '0'); | |
1605 | + break; | |
1606 | + case 10 ... 15: | |
1607 | + return (i - 10 + 'a'); | |
1608 | + break; | |
1609 | + default: | |
1610 | + if (net_ratelimit()) | |
1611 | + printk("layer7: Problem in dec2hex\n"); | |
1612 | + return '\0'; | |
1613 | + } | |
1614 | +} | |
1615 | + | |
1616 | +static char * hex_print(unsigned char * s) | |
1617 | +{ | |
1618 | + char * g = kmalloc(strlen(s)*3 + 1, GFP_ATOMIC); | |
1619 | + int i; | |
1620 | + | |
1621 | + if(!g) { | |
1622 | + if (net_ratelimit()) | |
1623 | + printk(KERN_ERR "layer7: out of memory in hex_print, " | |
1624 | + "bailing.\n"); | |
1625 | + return NULL; | |
1626 | + } | |
1627 | + | |
1628 | + for(i = 0; i < strlen(s); i++) { | |
1629 | + g[i*3 ] = dec2hex(s[i]/16); | |
1630 | + g[i*3 + 1] = dec2hex(s[i]%16); | |
1631 | + g[i*3 + 2] = ' '; | |
1632 | + } | |
1633 | + g[i*3] = '\0'; | |
1634 | + | |
1635 | + return g; | |
1636 | +} | |
1637 | +#endif // DEBUG | |
1638 | + | |
1639 | +/* Use instead of regcomp. As we expect to be seeing the same regexps over and | |
1640 | +over again, it make sense to cache the results. */ | |
1641 | +static regexp * compile_and_cache(const char * regex_string, | |
1642 | + const char * protocol) | |
1643 | +{ | |
1644 | + struct pattern_cache * node = first_pattern_cache; | |
1645 | + struct pattern_cache * last_pattern_cache = first_pattern_cache; | |
1646 | + struct pattern_cache * tmp; | |
1647 | + unsigned int len; | |
1648 | + | |
1649 | + while (node != NULL) { | |
1650 | + if (!strcmp(node->regex_string, regex_string)) | |
1651 | + return node->pattern; | |
1652 | + | |
1653 | + last_pattern_cache = node;/* points at the last non-NULL node */ | |
1654 | + node = node->next; | |
1655 | + } | |
1656 | + | |
1657 | + /* If we reach the end of the list, then we have not yet cached | |
1658 | + the pattern for this regex. Let's do that now. | |
1659 | + Be paranoid about running out of memory to avoid list corruption. */ | |
1660 | + tmp = kmalloc(sizeof(struct pattern_cache), GFP_ATOMIC); | |
1661 | + | |
1662 | + if(!tmp) { | |
1663 | + if (net_ratelimit()) | |
1664 | + printk(KERN_ERR "layer7: out of memory in " | |
1665 | + "compile_and_cache, bailing.\n"); | |
1666 | + return NULL; | |
1667 | + } | |
1668 | + | |
1669 | + tmp->regex_string = kmalloc(strlen(regex_string) + 1, GFP_ATOMIC); | |
1670 | + tmp->pattern = kmalloc(sizeof(struct regexp), GFP_ATOMIC); | |
1671 | + tmp->next = NULL; | |
1672 | + | |
1673 | + if(!tmp->regex_string || !tmp->pattern) { | |
1674 | + if (net_ratelimit()) | |
1675 | + printk(KERN_ERR "layer7: out of memory in " | |
1676 | + "compile_and_cache, bailing.\n"); | |
1677 | + kfree(tmp->regex_string); | |
1678 | + kfree(tmp->pattern); | |
1679 | + kfree(tmp); | |
1680 | + return NULL; | |
1681 | + } | |
1682 | + | |
1683 | + /* Ok. The new node is all ready now. */ | |
1684 | + node = tmp; | |
1685 | + | |
1686 | + if(first_pattern_cache == NULL) /* list is empty */ | |
1687 | + first_pattern_cache = node; /* make node the beginning */ | |
1688 | + else | |
1689 | + last_pattern_cache->next = node; /* attach node to the end */ | |
1690 | + | |
1691 | + /* copy the string and compile the regex */ | |
1692 | + len = strlen(regex_string); | |
1693 | + DPRINTK("About to compile this: \"%s\"\n", regex_string); | |
1694 | + node->pattern = regcomp((char *)regex_string, &len); | |
1695 | + if ( !node->pattern ) { | |
1696 | + if (net_ratelimit()) | |
1697 | + printk(KERN_ERR "layer7: Error compiling regexp " | |
1698 | + "\"%s\" (%s)\n", | |
1699 | + regex_string, protocol); | |
1700 | + /* pattern is now cached as NULL, so we won't try again. */ | |
1701 | + } | |
1702 | + | |
1703 | + strcpy(node->regex_string, regex_string); | |
1704 | + return node->pattern; | |
1705 | +} | |
1706 | + | |
1707 | +static int can_handle(const struct sk_buff *skb) | |
1708 | +{ | |
1709 | + struct iphdr iphdr_tmp; | |
1710 | + struct iphdr *iphdr; | |
1711 | + int offset; | |
1712 | + | |
1713 | + if (!ip_hdr(skb)) | |
1714 | + return 0; | |
1715 | + | |
1716 | + offset = ((uintptr_t)ip_hdr(skb)) - ((uintptr_t)skb->data); | |
1717 | + | |
1718 | + iphdr = skb_header_pointer(skb, offset, sizeof(*iphdr), &iphdr_tmp); | |
1719 | + if (!iphdr) | |
1720 | + return 0; | |
1721 | + | |
1722 | + if (iphdr->protocol == IPPROTO_TCP || | |
1723 | + iphdr->protocol == IPPROTO_UDP || | |
1724 | + iphdr->protocol == IPPROTO_ICMP) | |
1725 | + return 1; | |
1726 | + | |
1727 | + return 0; | |
1728 | +} | |
1729 | + | |
1730 | +static int app_data_offset(const struct sk_buff *skb) | |
1731 | +{ | |
1732 | + int offset; | |
1733 | + struct iphdr iphdr_tmp; | |
1734 | + struct iphdr *iphdr; | |
1735 | + struct tcphdr tcphdr_tmp; | |
1736 | + struct tcphdr *tcphdr; | |
1737 | + | |
1738 | + if (!ip_hdr(skb)) | |
1739 | + return -1; | |
1740 | + | |
1741 | + offset = ((uintptr_t)ip_hdr(skb)) - ((uintptr_t)skb->data); | |
1742 | + | |
1743 | + iphdr = skb_header_pointer(skb, offset, sizeof(*iphdr), &iphdr_tmp); | |
1744 | + if (!iphdr) | |
1745 | + return -1; | |
1746 | + | |
1747 | + offset += iphdr->ihl * 4; | |
1748 | + | |
1749 | + if (iphdr->protocol == IPPROTO_TCP) { | |
1750 | + tcphdr = skb_header_pointer(skb, offset, sizeof(*tcphdr), | |
1751 | + &tcphdr_tmp); | |
1752 | + if (!tcphdr) | |
1753 | + return -1; | |
1754 | + | |
1755 | + offset += tcphdr->doff * 4; | |
1756 | + | |
1757 | + return offset; | |
1758 | + } | |
1759 | + | |
1760 | + if (iphdr->protocol == IPPROTO_UDP) | |
1761 | + return offset + 8; | |
1762 | + | |
1763 | + if (iphdr->protocol == IPPROTO_ICMP) | |
1764 | + return offset + 8; | |
1765 | + | |
1766 | + if (net_ratelimit()) | |
1767 | + pr_err(KERN_ERR "layer7: tried to handle unknown protocol!\n"); | |
1768 | + | |
1769 | + return offset + 8; /* something reasonable */ | |
1770 | +} | |
1771 | + | |
1772 | +/* handles whether there's a match when we aren't appending data anymore */ | |
1773 | +static int match_no_append(struct nf_conn * conntrack, | |
1774 | + struct nf_conn * master_conntrack, | |
1775 | + enum ip_conntrack_info ctinfo, | |
1776 | + enum ip_conntrack_info master_ctinfo, | |
1777 | + const struct xt_layer7_info * info) | |
1778 | +{ | |
1779 | + /* If we're in here, throw the app data away */ | |
1780 | + if(master_conntrack->layer7.app_data != NULL) { | |
1781 | + | |
1782 | + #ifdef CONFIG_IP_NF_MATCH_LAYER7_DEBUG | |
1783 | + if(!master_conntrack->layer7.app_proto) { | |
1784 | + char * f = | |
1785 | + friendly_print(master_conntrack->layer7.app_data); | |
1786 | + char * g = | |
1787 | + hex_print(master_conntrack->layer7.app_data); | |
1788 | + DPRINTK("\nl7-filter gave up after %d bytes " | |
1789 | + "(%d packets):\n%s\n", | |
1790 | + strlen(f), total_acct_packets(master_conntrack), f); | |
1791 | + kfree(f); | |
1792 | + DPRINTK("In hex: %s\n", g); | |
1793 | + kfree(g); | |
1794 | + } | |
1795 | + #endif | |
1796 | + | |
1797 | + kfree(master_conntrack->layer7.app_data); | |
1798 | + master_conntrack->layer7.app_data = NULL; /* don't free again */ | |
1799 | + } | |
1800 | + | |
1801 | + if(master_conntrack->layer7.app_proto){ | |
1802 | + /* Here child connections set their .app_proto (for /proc) */ | |
1803 | + if(!conntrack->layer7.app_proto) { | |
1804 | + conntrack->layer7.app_proto = | |
1805 | + kmalloc(strlen(master_conntrack->layer7.app_proto)+1, | |
1806 | + GFP_ATOMIC); | |
1807 | + if(!conntrack->layer7.app_proto){ | |
1808 | + if (net_ratelimit()) | |
1809 | + printk(KERN_ERR "layer7: out of memory " | |
1810 | + "in match_no_append, " | |
1811 | + "bailing.\n"); | |
1812 | + return 1; | |
1813 | + } | |
1814 | + strcpy(conntrack->layer7.app_proto, | |
1815 | + master_conntrack->layer7.app_proto); | |
1816 | + } | |
1817 | + | |
1818 | + return (!strcmp(master_conntrack->layer7.app_proto, | |
1819 | + info->protocol)); | |
1820 | + } | |
1821 | + else { | |
1822 | + /* If not classified, set to "unknown" to distinguish from | |
1823 | + connections that are still being tested. */ | |
1824 | + master_conntrack->layer7.app_proto = | |
1825 | + kmalloc(strlen("unknown")+1, GFP_ATOMIC); | |
1826 | + if(!master_conntrack->layer7.app_proto){ | |
1827 | + if (net_ratelimit()) | |
1828 | + printk(KERN_ERR "layer7: out of memory in " | |
1829 | + "match_no_append, bailing.\n"); | |
1830 | + return 1; | |
1831 | + } | |
1832 | + strcpy(master_conntrack->layer7.app_proto, "unknown"); | |
1833 | + return 0; | |
1834 | + } | |
1835 | +} | |
1836 | + | |
1837 | +/* add the new app data to the conntrack. Return number of bytes added. */ | |
1838 | +static int add_datastr(char *target, int offset, char *app_data, int len) | |
1839 | +{ | |
1840 | + int length = 0, i; | |
1841 | + if (!target) return 0; | |
1842 | + | |
1843 | + /* Strip nulls. Make everything lower case (our regex lib doesn't | |
1844 | + do case insensitivity). Add it to the end of the current data. */ | |
1845 | + for(i = 0; i < maxdatalen-offset-1 && i < len; i++) { | |
1846 | + if(app_data[i] != '\0') { | |
1847 | + /* the kernel version of tolower mungs 'upper ascii' */ | |
1848 | + target[length+offset] = | |
1849 | + isascii(app_data[i])? | |
1850 | + tolower(app_data[i]) : app_data[i]; | |
1851 | + length++; | |
1852 | + } | |
1853 | + } | |
1854 | + target[length+offset] = '\0'; | |
1855 | + | |
1856 | + return length; | |
1857 | +} | |
1858 | + | |
1859 | +/* add the new app data to the buffer. Return number of bytes added. */ | |
1860 | +static int add_data(char *target, int offset, const struct sk_buff *skb) | |
1861 | +{ | |
1862 | + int length, length_sum = 0; | |
1863 | + int data_start = app_data_offset(skb); | |
1864 | + int remaining = skb->len - data_start; | |
1865 | + int to_copy; | |
1866 | + uint8_t buf[512]; | |
1867 | + uint8_t *data; | |
1868 | + | |
1869 | + while ((offset < maxdatalen - 1) && (remaining > 0)) { | |
1870 | + to_copy = min_t(int, remaining, sizeof(buf)); | |
1871 | + | |
1872 | + data = skb_header_pointer(skb, data_start, to_copy, buf); | |
1873 | + length = add_datastr(target, offset, data, to_copy); | |
1874 | + | |
1875 | + remaining -= to_copy; | |
1876 | + data_start += to_copy; | |
1877 | + offset += length; | |
1878 | + length_sum += length; | |
1879 | + } | |
1880 | + | |
1881 | + return length_sum; | |
1882 | +} | |
1883 | + | |
1884 | +/* add the new app data to the conntrack. Return number of bytes added. */ | |
1885 | +static int add_data_conntrack(struct nf_conn *master_conntrack, | |
1886 | + const struct sk_buff *skb) | |
1887 | +{ | |
1888 | + int length; | |
1889 | + | |
1890 | + length = add_data(master_conntrack->layer7.app_data, | |
1891 | + master_conntrack->layer7.app_data_len, skb); | |
1892 | + master_conntrack->layer7.app_data_len += length; | |
1893 | + | |
1894 | + return length; | |
1895 | +} | |
1896 | + | |
1897 | +/* taken from drivers/video/modedb.c */ | |
1898 | +static int my_atoi(const char *s) | |
1899 | +{ | |
1900 | + int val = 0; | |
1901 | + | |
1902 | + for (;; s++) { | |
1903 | + switch (*s) { | |
1904 | + case '0'...'9': | |
1905 | + val = 10*val+(*s-'0'); | |
1906 | + break; | |
1907 | + default: | |
1908 | + return val; | |
1909 | + } | |
1910 | + } | |
1911 | +} | |
1912 | + | |
1913 | +static int layer7_numpackets_proc_show(struct seq_file *s, void *p) { | |
1914 | + seq_printf(s, "%d\n", num_packets); | |
1915 | + | |
1916 | + return 0; | |
1917 | +} | |
1918 | + | |
1919 | +static int layer7_numpackets_proc_open(struct inode *inode, struct file *file) { | |
1920 | + return single_open(file, layer7_numpackets_proc_show, NULL); | |
1921 | +} | |
1922 | + | |
1923 | +/* Read in num_packets from userland */ | |
1924 | +static ssize_t layer7_numpackets_write_proc(struct file* file, const char __user *buffer, | |
1925 | + size_t count, loff_t *data) { | |
1926 | + char value[1024]; | |
1927 | + int new_num_packets; | |
1928 | + | |
1929 | + if (copy_from_user(&value, buffer, sizeof(value))) | |
1930 | + return -EFAULT; | |
1931 | + | |
1932 | + new_num_packets = my_atoi(value); | |
1933 | + | |
1934 | + if ((new_num_packets < 1) || (new_num_packets > 99)) { | |
1935 | + printk(KERN_WARNING "layer7: numpackets must be between 1 and 99\n"); | |
1936 | + return -EFAULT; | |
1937 | + } | |
1938 | + | |
1939 | + num_packets = new_num_packets; | |
1940 | + | |
1941 | + return count; | |
1942 | +} | |
1943 | + | |
1944 | +static bool match(const struct sk_buff *skbin, struct xt_action_param *par) | |
1945 | +{ | |
1946 | + /* sidestep const without getting a compiler warning... */ | |
1947 | + struct sk_buff *skb = (struct sk_buff *)skbin; | |
1948 | + | |
1949 | + const struct xt_layer7_info * info = par->matchinfo; | |
1950 | + | |
1951 | + enum ip_conntrack_info master_ctinfo, ctinfo; | |
1952 | + struct nf_conn *master_conntrack, *conntrack; | |
1953 | + unsigned char *tmp_data; | |
1954 | + unsigned int pattern_result; | |
1955 | + regexp * comppattern; | |
1956 | + | |
1957 | + /* Be paranoid/incompetent - lock the entire match function. */ | |
1958 | + spin_lock_bh(&l7_lock); | |
1959 | + | |
1960 | + if (!can_handle(skbin)) { | |
1961 | + DPRINTK("layer7: This is some protocol I can't handle.\n"); | |
1962 | + spin_unlock_bh(&l7_lock); | |
1963 | + return info->invert; | |
1964 | + } | |
1965 | + | |
1966 | + /* Treat parent & all its children together as one connection, except | |
1967 | + for the purpose of setting conntrack->layer7.app_proto in the actual | |
1968 | + connection. This makes /proc/net/ip_conntrack more satisfying. */ | |
1969 | + conntrack = nf_ct_get(skbin, &ctinfo); | |
1970 | + master_conntrack = nf_ct_get(skbin, &master_ctinfo); | |
1971 | + if (!conntrack || !master_conntrack) { | |
1972 | + DPRINTK("layer7: couldn't get conntrack.\n"); | |
1973 | + spin_unlock_bh(&l7_lock); | |
1974 | + return info->invert; | |
1975 | + } | |
1976 | + | |
1977 | + /* Try to get a master conntrack (and its master etc) for FTP, etc. */ | |
1978 | + while (master_ct(master_conntrack) != NULL) | |
1979 | + master_conntrack = master_ct(master_conntrack); | |
1980 | + | |
1981 | + /* if we've classified it or seen too many packets */ | |
1982 | + if(!info->pkt && (total_acct_packets(master_conntrack) > num_packets || | |
1983 | + master_conntrack->layer7.app_proto)) { | |
1984 | + | |
1985 | + pattern_result = match_no_append(conntrack, master_conntrack, | |
1986 | + ctinfo, master_ctinfo, info); | |
1987 | + | |
1988 | + /* skb->cb[0] == seen. Don't do things twice if there are | |
1989 | + multiple l7 rules. I'm not sure that using cb for this purpose | |
1990 | + is correct, even though it says "put your private variables | |
1991 | + there". But it doesn't look like it is being used for anything | |
1992 | + else in the skbs that make it here. */ | |
1993 | + skb->cb[0] = 1; /* marking it seen here's probably irrelevant */ | |
1994 | + | |
1995 | + spin_unlock_bh(&l7_lock); | |
1996 | + return (pattern_result ^ info->invert); | |
1997 | + } | |
1998 | + | |
1999 | + /* the return value gets checked later, when we're ready to use it */ | |
2000 | + comppattern = compile_and_cache(info->pattern, info->protocol); | |
2001 | + | |
2002 | + if (info->pkt) { | |
2003 | + tmp_data = kmalloc(maxdatalen, GFP_ATOMIC); | |
2004 | + if(!tmp_data){ | |
2005 | + if (net_ratelimit()) | |
2006 | + printk(KERN_ERR "layer7: out of memory in match, bailing.\n"); | |
2007 | + return info->invert; | |
2008 | + } | |
2009 | + | |
2010 | + tmp_data[0] = '\0'; | |
2011 | + add_data(tmp_data, 0, skbin); | |
2012 | + pattern_result = ((comppattern && regexec(comppattern, tmp_data)) ? 1 : 0); | |
2013 | + | |
2014 | + kfree(tmp_data); | |
2015 | + tmp_data = NULL; | |
2016 | + spin_unlock_bh(&l7_lock); | |
2017 | + | |
2018 | + return (pattern_result ^ info->invert); | |
2019 | + } | |
2020 | + | |
2021 | + /* On the first packet of a connection, allocate space for app data */ | |
2022 | + if(total_acct_packets(master_conntrack) == 1 && !skb->cb[0] && | |
2023 | + !master_conntrack->layer7.app_data){ | |
2024 | + master_conntrack->layer7.app_data = | |
2025 | + kmalloc(maxdatalen, GFP_ATOMIC); | |
2026 | + if(!master_conntrack->layer7.app_data){ | |
2027 | + if (net_ratelimit()) | |
2028 | + printk(KERN_ERR "layer7: out of memory in " | |
2029 | + "match, bailing.\n"); | |
2030 | + spin_unlock_bh(&l7_lock); | |
2031 | + return info->invert; | |
2032 | + } | |
2033 | + | |
2034 | + master_conntrack->layer7.app_data[0] = '\0'; | |
2035 | + } | |
2036 | + | |
2037 | + /* Can be here, but unallocated, if numpackets is increased near | |
2038 | + the beginning of a connection */ | |
2039 | + if(master_conntrack->layer7.app_data == NULL){ | |
2040 | + spin_unlock_bh(&l7_lock); | |
2041 | + return info->invert; /* unmatched */ | |
2042 | + } | |
2043 | + | |
2044 | + if(!skb->cb[0]){ | |
2045 | + int newbytes; | |
2046 | + newbytes = add_data_conntrack(master_conntrack, skb); | |
2047 | + | |
2048 | + if(newbytes == 0) { /* didn't add any data */ | |
2049 | + skb->cb[0] = 1; | |
2050 | + /* Didn't match before, not going to match now */ | |
2051 | + spin_unlock_bh(&l7_lock); | |
2052 | + return info->invert; | |
2053 | + } | |
2054 | + } | |
2055 | + | |
2056 | + /* If looking for "unknown", then never match. "Unknown" means that | |
2057 | + we've given up; we're still trying with these packets. */ | |
2058 | + if(!strcmp(info->protocol, "unknown")) { | |
2059 | + pattern_result = 0; | |
2060 | + /* If looking for "unset", then always match. "Unset" means that we | |
2061 | + haven't yet classified the connection. */ | |
2062 | + } else if(!strcmp(info->protocol, "unset")) { | |
2063 | + pattern_result = 2; | |
2064 | + DPRINTK("layer7: matched unset: not yet classified " | |
2065 | + "(%d/%d packets)\n", | |
2066 | + total_acct_packets(master_conntrack), num_packets); | |
2067 | + /* If the regexp failed to compile, don't bother running it */ | |
2068 | + } else if(comppattern && | |
2069 | + regexec(comppattern, master_conntrack->layer7.app_data)){ | |
2070 | + DPRINTK("layer7: matched %s\n", info->protocol); | |
2071 | + pattern_result = 1; | |
2072 | + } else pattern_result = 0; | |
2073 | + | |
2074 | + if(pattern_result == 1) { | |
2075 | + master_conntrack->layer7.app_proto = | |
2076 | + kmalloc(strlen(info->protocol)+1, GFP_ATOMIC); | |
2077 | + if(!master_conntrack->layer7.app_proto){ | |
2078 | + if (net_ratelimit()) | |
2079 | + printk(KERN_ERR "layer7: out of memory in " | |
2080 | + "match, bailing.\n"); | |
2081 | + spin_unlock_bh(&l7_lock); | |
2082 | + return (pattern_result ^ info->invert); | |
2083 | + } | |
2084 | + strcpy(master_conntrack->layer7.app_proto, info->protocol); | |
2085 | + } else if(pattern_result > 1) { /* cleanup from "unset" */ | |
2086 | + pattern_result = 1; | |
2087 | + } | |
2088 | + | |
2089 | + /* mark the packet seen */ | |
2090 | + skb->cb[0] = 1; | |
2091 | + | |
2092 | + spin_unlock_bh(&l7_lock); | |
2093 | + return (pattern_result ^ info->invert); | |
2094 | +} | |
2095 | + | |
2096 | +// load nf_conntrack_ipv4 | |
2097 | +static int check(const struct xt_mtchk_param *par) | |
2098 | +{ | |
2099 | + if (nf_ct_l3proto_try_module_get(par->match->family) < 0) { | |
2100 | + printk(KERN_WARNING "can't load conntrack support for " | |
2101 | + "proto=%d\n", par->match->family); | |
2102 | + return -EINVAL; | |
2103 | + } | |
2104 | + return 0; | |
2105 | +} | |
2106 | + | |
2107 | + | |
2108 | +static void destroy(const struct xt_mtdtor_param *par) | |
2109 | +{ | |
2110 | + nf_ct_l3proto_module_put(par->match->family); | |
2111 | +} | |
2112 | + | |
2113 | +static struct xt_match xt_layer7_match[] __read_mostly = { | |
2114 | +{ | |
2115 | + .name = "layer7", | |
2116 | + .family = AF_INET, | |
2117 | + .checkentry = check, | |
2118 | + .match = match, | |
2119 | + .destroy = destroy, | |
2120 | + .matchsize = sizeof(struct xt_layer7_info), | |
2121 | + .me = THIS_MODULE | |
2122 | +} | |
2123 | +}; | |
2124 | + | |
2125 | +static const struct file_operations layer7_numpackets_proc_fops = { | |
2126 | + .owner = THIS_MODULE, | |
2127 | + .open = layer7_numpackets_proc_open, | |
2128 | + .read = seq_read, | |
2129 | + .llseek = seq_lseek, | |
2130 | + .release = single_release, | |
2131 | + .write = layer7_numpackets_write_proc, | |
2132 | +}; | |
2133 | + | |
2134 | +static int __init xt_layer7_init(void) | |
2135 | +{ | |
2136 | + need_conntrack(); | |
2137 | + | |
2138 | + // Register proc interface | |
2139 | + proc_create_data("layer7_numpackets", 0644, | |
2140 | + init_net.proc_net, &layer7_numpackets_proc_fops, NULL); | |
2141 | + | |
2142 | + if(maxdatalen < 1) { | |
2143 | + printk(KERN_WARNING "layer7: maxdatalen can't be < 1, " | |
2144 | + "using 1\n"); | |
2145 | + maxdatalen = 1; | |
2146 | + } | |
2147 | + /* This is not a hard limit. It's just here to prevent people from | |
2148 | + bringing their slow machines to a grinding halt. */ | |
2149 | + else if(maxdatalen > 65536) { | |
2150 | + printk(KERN_WARNING "layer7: maxdatalen can't be > 65536, " | |
2151 | + "using 65536\n"); | |
2152 | + maxdatalen = 65536; | |
2153 | + } | |
2154 | + return xt_register_matches(xt_layer7_match, | |
2155 | + ARRAY_SIZE(xt_layer7_match)); | |
2156 | +} | |
2157 | + | |
2158 | +static void __exit xt_layer7_fini(void) | |
2159 | +{ | |
2160 | + remove_proc_entry("layer7_numpackets", init_net.proc_net); | |
2161 | + xt_unregister_matches(xt_layer7_match, ARRAY_SIZE(xt_layer7_match)); | |
2162 | +} | |
2163 | + | |
2164 | +module_init(xt_layer7_init); | |
2165 | +module_exit(xt_layer7_fini); |