kernel: Update layer7 patch.
[people/teissler/ipfire-2.x.git] / src / patches / linux-3.10-layer7-filter.patch
1 http://git.ipfire.org/?p=people/ms/linux.git;a=shortlog;h=refs/heads/linux-3.10.y-layer7
2
3 diff --git a/include/linux/netfilter/xt_layer7.h b/include/linux/netfilter/xt_layer7.h
4 new file mode 100644
5 index 0000000..c38d3c4
6 --- /dev/null
7 +++ b/include/linux/netfilter/xt_layer7.h
8 @@ -0,0 +1,14 @@
9 +#ifndef _XT_LAYER7_H
10 +#define _XT_LAYER7_H
11 +
12 +#define MAX_PATTERN_LEN 8192
13 +#define MAX_PROTOCOL_LEN 256
14 +
15 +struct xt_layer7_info {
16 + char protocol[MAX_PROTOCOL_LEN];
17 + char pattern[MAX_PATTERN_LEN];
18 + u_int8_t invert;
19 + u_int8_t pkt;
20 +};
21 +
22 +#endif /* _XT_LAYER7_H */
23 diff --git a/include/net/netfilter/nf_conntrack.h b/include/net/netfilter/nf_conntrack.h
24 index 644d9c2..b226a8f 100644
25 --- a/include/net/netfilter/nf_conntrack.h
26 +++ b/include/net/netfilter/nf_conntrack.h
27 @@ -105,6 +105,22 @@ struct nf_conn {
28 struct net *ct_net;
29 #endif
30
31 +#if defined(CONFIG_NETFILTER_XT_MATCH_LAYER7) || \
32 + defined(CONFIG_NETFILTER_XT_MATCH_LAYER7_MODULE)
33 + struct {
34 + /*
35 + * e.g. "http". NULL before decision. "unknown" after decision
36 + * if no match.
37 + */
38 + char *app_proto;
39 + /*
40 + * application layer data so far. NULL after match decision.
41 + */
42 + char *app_data;
43 + unsigned int app_data_len;
44 + } layer7;
45 +#endif
46 +
47 /* Storage reserved for other modules, must be the last member */
48 union nf_conntrack_proto proto;
49 };
50 diff --git a/include/uapi/linux/netfilter/Kbuild b/include/uapi/linux/netfilter/Kbuild
51 index 4111577..a95e6b5 100644
52 --- a/include/uapi/linux/netfilter/Kbuild
53 +++ b/include/uapi/linux/netfilter/Kbuild
54 @@ -53,6 +53,7 @@ header-y += xt_hashlimit.h
55 header-y += xt_helper.h
56 header-y += xt_iprange.h
57 header-y += xt_ipvs.h
58 +header-y += xt_layer7.h
59 header-y += xt_length.h
60 header-y += xt_limit.h
61 header-y += xt_mac.h
62 diff --git a/net/netfilter/Kconfig b/net/netfilter/Kconfig
63 index 56d22ca..7827ba4 100644
64 --- a/net/netfilter/Kconfig
65 +++ b/net/netfilter/Kconfig
66 @@ -1011,6 +1011,26 @@ config NETFILTER_XT_MATCH_IPVS
67
68 If unsure, say N.
69
70 +config NETFILTER_XT_MATCH_LAYER7
71 + tristate '"layer7" match support'
72 + depends on NETFILTER_XTABLES
73 + depends on NETFILTER_ADVANCED
74 + depends on NF_CONNTRACK
75 + help
76 + Say Y if you want to be able to classify connections (and their
77 + packets) based on regular expression matching of their application
78 + layer data. This is one way to classify applications such as
79 + peer-to-peer filesharing systems that do not always use the same
80 + port.
81 +
82 + To compile it as a module, choose M here. If unsure, say N.
83 +
84 +config NETFILTER_XT_MATCH_LAYER7_DEBUG
85 + bool 'Layer 7 debugging output'
86 + depends on NETFILTER_XT_MATCH_LAYER7
87 + help
88 + Say Y to get lots of debugging output.
89 +
90 config NETFILTER_XT_MATCH_LENGTH
91 tristate '"length" match support'
92 depends on NETFILTER_ADVANCED
93 @@ -1205,6 +1225,12 @@ config NETFILTER_XT_MATCH_STATE
94
95 To compile it as a module, choose M here. If unsure, say N.
96
97 +config NETFILTER_XT_MATCH_LAYER7_DEBUG
98 + bool 'Layer 7 debugging output'
99 + depends on NETFILTER_XT_MATCH_LAYER7
100 + help
101 + Say Y to get lots of debugging output.
102 +
103 config NETFILTER_XT_MATCH_STATISTIC
104 tristate '"statistic" match support'
105 depends on NETFILTER_ADVANCED
106 diff --git a/net/netfilter/Makefile b/net/netfilter/Makefile
107 index a1abf87..acec24e 100644
108 --- a/net/netfilter/Makefile
109 +++ b/net/netfilter/Makefile
110 @@ -134,6 +134,7 @@ obj-$(CONFIG_NETFILTER_XT_MATCH_RECENT) += xt_recent.o
111 obj-$(CONFIG_NETFILTER_XT_MATCH_SCTP) += xt_sctp.o
112 obj-$(CONFIG_NETFILTER_XT_MATCH_SOCKET) += xt_socket.o
113 obj-$(CONFIG_NETFILTER_XT_MATCH_STATE) += xt_state.o
114 +obj-$(CONFIG_NETFILTER_XT_MATCH_LAYER7) += xt_layer7.o
115 obj-$(CONFIG_NETFILTER_XT_MATCH_STATISTIC) += xt_statistic.o
116 obj-$(CONFIG_NETFILTER_XT_MATCH_STRING) += xt_string.o
117 obj-$(CONFIG_NETFILTER_XT_MATCH_TCPMSS) += xt_tcpmss.o
118 diff --git a/net/netfilter/nf_conntrack_core.c b/net/netfilter/nf_conntrack_core.c
119 index 0283bae..cae3790 100644
120 --- a/net/netfilter/nf_conntrack_core.c
121 +++ b/net/netfilter/nf_conntrack_core.c
122 @@ -224,6 +224,13 @@ destroy_conntrack(struct nf_conntrack *nfct)
123 * too. */
124 nf_ct_remove_expectations(ct);
125
126 +#if defined(CONFIG_NETFILTER_XT_MATCH_LAYER7) || defined(CONFIG_NETFILTER_XT_MATCH_LAYER7_MODULE)
127 + if(ct->layer7.app_proto)
128 + kfree(ct->layer7.app_proto);
129 + if(ct->layer7.app_data)
130 + kfree(ct->layer7.app_data);
131 +#endif
132 +
133 /* We overload first tuple to link into unconfirmed or dying list.*/
134 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
135 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
136 diff --git a/net/netfilter/nf_conntrack_standalone.c b/net/netfilter/nf_conntrack_standalone.c
137 index bd700b4..5d045ca 100644
138 --- a/net/netfilter/nf_conntrack_standalone.c
139 +++ b/net/netfilter/nf_conntrack_standalone.c
140 @@ -240,6 +240,12 @@ static int ct_seq_show(struct seq_file *s, void *v)
141 if (ct_show_delta_time(s, ct))
142 goto release;
143
144 +#if defined(CONFIG_NETFILTER_XT_MATCH_LAYER7) || defined(CONFIG_NETFILTER_XT_MATCH_LAYER7_MODULE)
145 + if(ct->layer7.app_proto &&
146 + seq_printf(s, "l7proto=%s ", ct->layer7.app_proto))
147 + return -ENOSPC;
148 +#endif
149 +
150 if (seq_printf(s, "use=%u\n", atomic_read(&ct->ct_general.use)))
151 goto release;
152
153 diff --git a/net/netfilter/regexp/regexp.c b/net/netfilter/regexp/regexp.c
154 new file mode 100644
155 index 0000000..9006988
156 --- /dev/null
157 +++ b/net/netfilter/regexp/regexp.c
158 @@ -0,0 +1,1197 @@
159 +/*
160 + * regcomp and regexec -- regsub and regerror are elsewhere
161 + * @(#)regexp.c 1.3 of 18 April 87
162 + *
163 + * Copyright (c) 1986 by University of Toronto.
164 + * Written by Henry Spencer. Not derived from licensed software.
165 + *
166 + * Permission is granted to anyone to use this software for any
167 + * purpose on any computer system, and to redistribute it freely,
168 + * subject to the following restrictions:
169 + *
170 + * 1. The author is not responsible for the consequences of use of
171 + * this software, no matter how awful, even if they arise
172 + * from defects in it.
173 + *
174 + * 2. The origin of this software must not be misrepresented, either
175 + * by explicit claim or by omission.
176 + *
177 + * 3. Altered versions must be plainly marked as such, and must not
178 + * be misrepresented as being the original software.
179 + *
180 + * Beware that some of this code is subtly aware of the way operator
181 + * precedence is structured in regular expressions. Serious changes in
182 + * regular-expression syntax might require a total rethink.
183 + *
184 + * This code was modified by Ethan Sommer to work within the kernel
185 + * (it now uses kmalloc etc..)
186 + *
187 + * Modified slightly by Matthew Strait to use more modern C.
188 + */
189 +
190 +#include "regexp.h"
191 +#include "regmagic.h"
192 +
193 +/* added by ethan and matt. Lets it work in both kernel and user space.
194 +(So iptables can use it, for instance.) Yea, it goes both ways... */
195 +#if __KERNEL__
196 + #define malloc(foo) kmalloc(foo,GFP_ATOMIC)
197 +#else
198 + #define printk(format,args...) printf(format,##args)
199 +#endif
200 +
201 +void regerror(char * s)
202 +{
203 + printk("<3>Regexp: %s\n", s);
204 + /* NOTREACHED */
205 +}
206 +
207 +/*
208 + * The "internal use only" fields in regexp.h are present to pass info from
209 + * compile to execute that permits the execute phase to run lots faster on
210 + * simple cases. They are:
211 + *
212 + * regstart char that must begin a match; '\0' if none obvious
213 + * reganch is the match anchored (at beginning-of-line only)?
214 + * regmust string (pointer into program) that match must include, or NULL
215 + * regmlen length of regmust string
216 + *
217 + * Regstart and reganch permit very fast decisions on suitable starting points
218 + * for a match, cutting down the work a lot. Regmust permits fast rejection
219 + * of lines that cannot possibly match. The regmust tests are costly enough
220 + * that regcomp() supplies a regmust only if the r.e. contains something
221 + * potentially expensive (at present, the only such thing detected is * or +
222 + * at the start of the r.e., which can involve a lot of backup). Regmlen is
223 + * supplied because the test in regexec() needs it and regcomp() is computing
224 + * it anyway.
225 + */
226 +
227 +/*
228 + * Structure for regexp "program". This is essentially a linear encoding
229 + * of a nondeterministic finite-state machine (aka syntax charts or
230 + * "railroad normal form" in parsing technology). Each node is an opcode
231 + * plus a "next" pointer, possibly plus an operand. "Next" pointers of
232 + * all nodes except BRANCH implement concatenation; a "next" pointer with
233 + * a BRANCH on both ends of it is connecting two alternatives. (Here we
234 + * have one of the subtle syntax dependencies: an individual BRANCH (as
235 + * opposed to a collection of them) is never concatenated with anything
236 + * because of operator precedence.) The operand of some types of node is
237 + * a literal string; for others, it is a node leading into a sub-FSM. In
238 + * particular, the operand of a BRANCH node is the first node of the branch.
239 + * (NB this is *not* a tree structure: the tail of the branch connects
240 + * to the thing following the set of BRANCHes.) The opcodes are:
241 + */
242 +
243 +/* definition number opnd? meaning */
244 +#define END 0 /* no End of program. */
245 +#define BOL 1 /* no Match "" at beginning of line. */
246 +#define EOL 2 /* no Match "" at end of line. */
247 +#define ANY 3 /* no Match any one character. */
248 +#define ANYOF 4 /* str Match any character in this string. */
249 +#define ANYBUT 5 /* str Match any character not in this string. */
250 +#define BRANCH 6 /* node Match this alternative, or the next... */
251 +#define BACK 7 /* no Match "", "next" ptr points backward. */
252 +#define EXACTLY 8 /* str Match this string. */
253 +#define NOTHING 9 /* no Match empty string. */
254 +#define STAR 10 /* node Match this (simple) thing 0 or more times. */
255 +#define PLUS 11 /* node Match this (simple) thing 1 or more times. */
256 +#define OPEN 20 /* no Mark this point in input as start of #n. */
257 + /* OPEN+1 is number 1, etc. */
258 +#define CLOSE 30 /* no Analogous to OPEN. */
259 +
260 +/*
261 + * Opcode notes:
262 + *
263 + * BRANCH The set of branches constituting a single choice are hooked
264 + * together with their "next" pointers, since precedence prevents
265 + * anything being concatenated to any individual branch. The
266 + * "next" pointer of the last BRANCH in a choice points to the
267 + * thing following the whole choice. This is also where the
268 + * final "next" pointer of each individual branch points; each
269 + * branch starts with the operand node of a BRANCH node.
270 + *
271 + * BACK Normal "next" pointers all implicitly point forward; BACK
272 + * exists to make loop structures possible.
273 + *
274 + * STAR,PLUS '?', and complex '*' and '+', are implemented as circular
275 + * BRANCH structures using BACK. Simple cases (one character
276 + * per match) are implemented with STAR and PLUS for speed
277 + * and to minimize recursive plunges.
278 + *
279 + * OPEN,CLOSE ...are numbered at compile time.
280 + */
281 +
282 +/*
283 + * A node is one char of opcode followed by two chars of "next" pointer.
284 + * "Next" pointers are stored as two 8-bit pieces, high order first. The
285 + * value is a positive offset from the opcode of the node containing it.
286 + * An operand, if any, simply follows the node. (Note that much of the
287 + * code generation knows about this implicit relationship.)
288 + *
289 + * Using two bytes for the "next" pointer is vast overkill for most things,
290 + * but allows patterns to get big without disasters.
291 + */
292 +#define OP(p) (*(p))
293 +#define NEXT(p) (((*((p)+1)&0377)<<8) + (*((p)+2)&0377))
294 +#define OPERAND(p) ((p) + 3)
295 +
296 +/*
297 + * See regmagic.h for one further detail of program structure.
298 + */
299 +
300 +
301 +/*
302 + * Utility definitions.
303 + */
304 +#ifndef CHARBITS
305 +#define UCHARAT(p) ((int)*(unsigned char *)(p))
306 +#else
307 +#define UCHARAT(p) ((int)*(p)&CHARBITS)
308 +#endif
309 +
310 +#define FAIL(m) { regerror(m); return(NULL); }
311 +#define ISMULT(c) ((c) == '*' || (c) == '+' || (c) == '?')
312 +#define META "^$.[()|?+*\\"
313 +
314 +/*
315 + * Flags to be passed up and down.
316 + */
317 +#define HASWIDTH 01 /* Known never to match null string. */
318 +#define SIMPLE 02 /* Simple enough to be STAR/PLUS operand. */
319 +#define SPSTART 04 /* Starts with * or +. */
320 +#define WORST 0 /* Worst case. */
321 +
322 +/*
323 + * Global work variables for regcomp().
324 + */
325 +struct match_globals {
326 +char *reginput; /* String-input pointer. */
327 +char *regbol; /* Beginning of input, for ^ check. */
328 +char **regstartp; /* Pointer to startp array. */
329 +char **regendp; /* Ditto for endp. */
330 +char *regparse; /* Input-scan pointer. */
331 +int regnpar; /* () count. */
332 +char regdummy;
333 +char *regcode; /* Code-emit pointer; &regdummy = don't. */
334 +long regsize; /* Code size. */
335 +};
336 +
337 +/*
338 + * Forward declarations for regcomp()'s friends.
339 + */
340 +#ifndef STATIC
341 +#define STATIC static
342 +#endif
343 +STATIC char *reg(struct match_globals *g, int paren,int *flagp);
344 +STATIC char *regbranch(struct match_globals *g, int *flagp);
345 +STATIC char *regpiece(struct match_globals *g, int *flagp);
346 +STATIC char *regatom(struct match_globals *g, int *flagp);
347 +STATIC char *regnode(struct match_globals *g, char op);
348 +STATIC char *regnext(struct match_globals *g, char *p);
349 +STATIC void regc(struct match_globals *g, char b);
350 +STATIC void reginsert(struct match_globals *g, char op, char *opnd);
351 +STATIC void regtail(struct match_globals *g, char *p, char *val);
352 +STATIC void regoptail(struct match_globals *g, char *p, char *val);
353 +
354 +
355 +__kernel_size_t my_strcspn(const char *s1,const char *s2)
356 +{
357 + char *scan1;
358 + char *scan2;
359 + int count;
360 +
361 + count = 0;
362 + for (scan1 = (char *)s1; *scan1 != '\0'; scan1++) {
363 + for (scan2 = (char *)s2; *scan2 != '\0';) /* ++ moved down. */
364 + if (*scan1 == *scan2++)
365 + return(count);
366 + count++;
367 + }
368 + return(count);
369 +}
370 +
371 +/*
372 + - regcomp - compile a regular expression into internal code
373 + *
374 + * We can't allocate space until we know how big the compiled form will be,
375 + * but we can't compile it (and thus know how big it is) until we've got a
376 + * place to put the code. So we cheat: we compile it twice, once with code
377 + * generation turned off and size counting turned on, and once "for real".
378 + * This also means that we don't allocate space until we are sure that the
379 + * thing really will compile successfully, and we never have to move the
380 + * code and thus invalidate pointers into it. (Note that it has to be in
381 + * one piece because free() must be able to free it all.)
382 + *
383 + * Beware that the optimization-preparation code in here knows about some
384 + * of the structure of the compiled regexp.
385 + */
386 +regexp *
387 +regcomp(char *exp,int *patternsize)
388 +{
389 + register regexp *r;
390 + register char *scan;
391 + register char *longest;
392 + register int len;
393 + int flags;
394 + struct match_globals g;
395 +
396 + /* commented out by ethan
397 + extern char *malloc();
398 + */
399 +
400 + if (exp == NULL)
401 + FAIL("NULL argument");
402 +
403 + /* First pass: determine size, legality. */
404 + g.regparse = exp;
405 + g.regnpar = 1;
406 + g.regsize = 0L;
407 + g.regcode = &g.regdummy;
408 + regc(&g, MAGIC);
409 + if (reg(&g, 0, &flags) == NULL)
410 + return(NULL);
411 +
412 + /* Small enough for pointer-storage convention? */
413 + if (g.regsize >= 32767L) /* Probably could be 65535L. */
414 + FAIL("regexp too big");
415 +
416 + /* Allocate space. */
417 + *patternsize=sizeof(regexp) + (unsigned)g.regsize;
418 + r = (regexp *)malloc(sizeof(regexp) + (unsigned)g.regsize);
419 + if (r == NULL)
420 + FAIL("out of space");
421 +
422 + /* Second pass: emit code. */
423 + g.regparse = exp;
424 + g.regnpar = 1;
425 + g.regcode = r->program;
426 + regc(&g, MAGIC);
427 + if (reg(&g, 0, &flags) == NULL)
428 + return(NULL);
429 +
430 + /* Dig out information for optimizations. */
431 + r->regstart = '\0'; /* Worst-case defaults. */
432 + r->reganch = 0;
433 + r->regmust = NULL;
434 + r->regmlen = 0;
435 + scan = r->program+1; /* First BRANCH. */
436 + if (OP(regnext(&g, scan)) == END) { /* Only one top-level choice. */
437 + scan = OPERAND(scan);
438 +
439 + /* Starting-point info. */
440 + if (OP(scan) == EXACTLY)
441 + r->regstart = *OPERAND(scan);
442 + else if (OP(scan) == BOL)
443 + r->reganch++;
444 +
445 + /*
446 + * If there's something expensive in the r.e., find the
447 + * longest literal string that must appear and make it the
448 + * regmust. Resolve ties in favor of later strings, since
449 + * the regstart check works with the beginning of the r.e.
450 + * and avoiding duplication strengthens checking. Not a
451 + * strong reason, but sufficient in the absence of others.
452 + */
453 + if (flags&SPSTART) {
454 + longest = NULL;
455 + len = 0;
456 + for (; scan != NULL; scan = regnext(&g, scan))
457 + if (OP(scan) == EXACTLY && strlen(OPERAND(scan)) >= len) {
458 + longest = OPERAND(scan);
459 + len = strlen(OPERAND(scan));
460 + }
461 + r->regmust = longest;
462 + r->regmlen = len;
463 + }
464 + }
465 +
466 + return(r);
467 +}
468 +
469 +/*
470 + - reg - regular expression, i.e. main body or parenthesized thing
471 + *
472 + * Caller must absorb opening parenthesis.
473 + *
474 + * Combining parenthesis handling with the base level of regular expression
475 + * is a trifle forced, but the need to tie the tails of the branches to what
476 + * follows makes it hard to avoid.
477 + */
478 +static char *
479 +reg(struct match_globals *g, int paren, int *flagp /* Parenthesized? */ )
480 +{
481 + register char *ret;
482 + register char *br;
483 + register char *ender;
484 + register int parno = 0; /* 0 makes gcc happy */
485 + int flags;
486 +
487 + *flagp = HASWIDTH; /* Tentatively. */
488 +
489 + /* Make an OPEN node, if parenthesized. */
490 + if (paren) {
491 + if (g->regnpar >= NSUBEXP)
492 + FAIL("too many ()");
493 + parno = g->regnpar;
494 + g->regnpar++;
495 + ret = regnode(g, OPEN+parno);
496 + } else
497 + ret = NULL;
498 +
499 + /* Pick up the branches, linking them together. */
500 + br = regbranch(g, &flags);
501 + if (br == NULL)
502 + return(NULL);
503 + if (ret != NULL)
504 + regtail(g, ret, br); /* OPEN -> first. */
505 + else
506 + ret = br;
507 + if (!(flags&HASWIDTH))
508 + *flagp &= ~HASWIDTH;
509 + *flagp |= flags&SPSTART;
510 + while (*g->regparse == '|') {
511 + g->regparse++;
512 + br = regbranch(g, &flags);
513 + if (br == NULL)
514 + return(NULL);
515 + regtail(g, ret, br); /* BRANCH -> BRANCH. */
516 + if (!(flags&HASWIDTH))
517 + *flagp &= ~HASWIDTH;
518 + *flagp |= flags&SPSTART;
519 + }
520 +
521 + /* Make a closing node, and hook it on the end. */
522 + ender = regnode(g, (paren) ? CLOSE+parno : END);
523 + regtail(g, ret, ender);
524 +
525 + /* Hook the tails of the branches to the closing node. */
526 + for (br = ret; br != NULL; br = regnext(g, br))
527 + regoptail(g, br, ender);
528 +
529 + /* Check for proper termination. */
530 + if (paren && *g->regparse++ != ')') {
531 + FAIL("unmatched ()");
532 + } else if (!paren && *g->regparse != '\0') {
533 + if (*g->regparse == ')') {
534 + FAIL("unmatched ()");
535 + } else
536 + FAIL("junk on end"); /* "Can't happen". */
537 + /* NOTREACHED */
538 + }
539 +
540 + return(ret);
541 +}
542 +
543 +/*
544 + - regbranch - one alternative of an | operator
545 + *
546 + * Implements the concatenation operator.
547 + */
548 +static char *
549 +regbranch(struct match_globals *g, int *flagp)
550 +{
551 + register char *ret;
552 + register char *chain;
553 + register char *latest;
554 + int flags;
555 +
556 + *flagp = WORST; /* Tentatively. */
557 +
558 + ret = regnode(g, BRANCH);
559 + chain = NULL;
560 + while (*g->regparse != '\0' && *g->regparse != '|' && *g->regparse != ')') {
561 + latest = regpiece(g, &flags);
562 + if (latest == NULL)
563 + return(NULL);
564 + *flagp |= flags&HASWIDTH;
565 + if (chain == NULL) /* First piece. */
566 + *flagp |= flags&SPSTART;
567 + else
568 + regtail(g, chain, latest);
569 + chain = latest;
570 + }
571 + if (chain == NULL) /* Loop ran zero times. */
572 + (void) regnode(g, NOTHING);
573 +
574 + return(ret);
575 +}
576 +
577 +/*
578 + - regpiece - something followed by possible [*+?]
579 + *
580 + * Note that the branching code sequences used for ? and the general cases
581 + * of * and + are somewhat optimized: they use the same NOTHING node as
582 + * both the endmarker for their branch list and the body of the last branch.
583 + * It might seem that this node could be dispensed with entirely, but the
584 + * endmarker role is not redundant.
585 + */
586 +static char *
587 +regpiece(struct match_globals *g, int *flagp)
588 +{
589 + register char *ret;
590 + register char op;
591 + register char *next;
592 + int flags;
593 +
594 + ret = regatom(g, &flags);
595 + if (ret == NULL)
596 + return(NULL);
597 +
598 + op = *g->regparse;
599 + if (!ISMULT(op)) {
600 + *flagp = flags;
601 + return(ret);
602 + }
603 +
604 + if (!(flags&HASWIDTH) && op != '?')
605 + FAIL("*+ operand could be empty");
606 + *flagp = (op != '+') ? (WORST|SPSTART) : (WORST|HASWIDTH);
607 +
608 + if (op == '*' && (flags&SIMPLE))
609 + reginsert(g, STAR, ret);
610 + else if (op == '*') {
611 + /* Emit x* as (x&|), where & means "self". */
612 + reginsert(g, BRANCH, ret); /* Either x */
613 + regoptail(g, ret, regnode(g, BACK)); /* and loop */
614 + regoptail(g, ret, ret); /* back */
615 + regtail(g, ret, regnode(g, BRANCH)); /* or */
616 + regtail(g, ret, regnode(g, NOTHING)); /* null. */
617 + } else if (op == '+' && (flags&SIMPLE))
618 + reginsert(g, PLUS, ret);
619 + else if (op == '+') {
620 + /* Emit x+ as x(&|), where & means "self". */
621 + next = regnode(g, BRANCH); /* Either */
622 + regtail(g, ret, next);
623 + regtail(g, regnode(g, BACK), ret); /* loop back */
624 + regtail(g, next, regnode(g, BRANCH)); /* or */
625 + regtail(g, ret, regnode(g, NOTHING)); /* null. */
626 + } else if (op == '?') {
627 + /* Emit x? as (x|) */
628 + reginsert(g, BRANCH, ret); /* Either x */
629 + regtail(g, ret, regnode(g, BRANCH)); /* or */
630 + next = regnode(g, NOTHING); /* null. */
631 + regtail(g, ret, next);
632 + regoptail(g, ret, next);
633 + }
634 + g->regparse++;
635 + if (ISMULT(*g->regparse))
636 + FAIL("nested *?+");
637 +
638 + return(ret);
639 +}
640 +
641 +/*
642 + - regatom - the lowest level
643 + *
644 + * Optimization: gobbles an entire sequence of ordinary characters so that
645 + * it can turn them into a single node, which is smaller to store and
646 + * faster to run. Backslashed characters are exceptions, each becoming a
647 + * separate node; the code is simpler that way and it's not worth fixing.
648 + */
649 +static char *
650 +regatom(struct match_globals *g, int *flagp)
651 +{
652 + register char *ret;
653 + int flags;
654 +
655 + *flagp = WORST; /* Tentatively. */
656 +
657 + switch (*g->regparse++) {
658 + case '^':
659 + ret = regnode(g, BOL);
660 + break;
661 + case '$':
662 + ret = regnode(g, EOL);
663 + break;
664 + case '.':
665 + ret = regnode(g, ANY);
666 + *flagp |= HASWIDTH|SIMPLE;
667 + break;
668 + case '[': {
669 + register int class;
670 + register int classend;
671 +
672 + if (*g->regparse == '^') { /* Complement of range. */
673 + ret = regnode(g, ANYBUT);
674 + g->regparse++;
675 + } else
676 + ret = regnode(g, ANYOF);
677 + if (*g->regparse == ']' || *g->regparse == '-')
678 + regc(g, *g->regparse++);
679 + while (*g->regparse != '\0' && *g->regparse != ']') {
680 + if (*g->regparse == '-') {
681 + g->regparse++;
682 + if (*g->regparse == ']' || *g->regparse == '\0')
683 + regc(g, '-');
684 + else {
685 + class = UCHARAT(g->regparse-2)+1;
686 + classend = UCHARAT(g->regparse);
687 + if (class > classend+1)
688 + FAIL("invalid [] range");
689 + for (; class <= classend; class++)
690 + regc(g, class);
691 + g->regparse++;
692 + }
693 + } else
694 + regc(g, *g->regparse++);
695 + }
696 + regc(g, '\0');
697 + if (*g->regparse != ']')
698 + FAIL("unmatched []");
699 + g->regparse++;
700 + *flagp |= HASWIDTH|SIMPLE;
701 + }
702 + break;
703 + case '(':
704 + ret = reg(g, 1, &flags);
705 + if (ret == NULL)
706 + return(NULL);
707 + *flagp |= flags&(HASWIDTH|SPSTART);
708 + break;
709 + case '\0':
710 + case '|':
711 + case ')':
712 + FAIL("internal urp"); /* Supposed to be caught earlier. */
713 + break;
714 + case '?':
715 + case '+':
716 + case '*':
717 + FAIL("?+* follows nothing");
718 + break;
719 + case '\\':
720 + if (*g->regparse == '\0')
721 + FAIL("trailing \\");
722 + ret = regnode(g, EXACTLY);
723 + regc(g, *g->regparse++);
724 + regc(g, '\0');
725 + *flagp |= HASWIDTH|SIMPLE;
726 + break;
727 + default: {
728 + register int len;
729 + register char ender;
730 +
731 + g->regparse--;
732 + len = my_strcspn((const char *)g->regparse, (const char *)META);
733 + if (len <= 0)
734 + FAIL("internal disaster");
735 + ender = *(g->regparse+len);
736 + if (len > 1 && ISMULT(ender))
737 + len--; /* Back off clear of ?+* operand. */
738 + *flagp |= HASWIDTH;
739 + if (len == 1)
740 + *flagp |= SIMPLE;
741 + ret = regnode(g, EXACTLY);
742 + while (len > 0) {
743 + regc(g, *g->regparse++);
744 + len--;
745 + }
746 + regc(g, '\0');
747 + }
748 + break;
749 + }
750 +
751 + return(ret);
752 +}
753 +
754 +/*
755 + - regnode - emit a node
756 + */
757 +static char * /* Location. */
758 +regnode(struct match_globals *g, char op)
759 +{
760 + register char *ret;
761 + register char *ptr;
762 +
763 + ret = g->regcode;
764 + if (ret == &g->regdummy) {
765 + g->regsize += 3;
766 + return(ret);
767 + }
768 +
769 + ptr = ret;
770 + *ptr++ = op;
771 + *ptr++ = '\0'; /* Null "next" pointer. */
772 + *ptr++ = '\0';
773 + g->regcode = ptr;
774 +
775 + return(ret);
776 +}
777 +
778 +/*
779 + - regc - emit (if appropriate) a byte of code
780 + */
781 +static void
782 +regc(struct match_globals *g, char b)
783 +{
784 + if (g->regcode != &g->regdummy)
785 + *g->regcode++ = b;
786 + else
787 + g->regsize++;
788 +}
789 +
790 +/*
791 + - reginsert - insert an operator in front of already-emitted operand
792 + *
793 + * Means relocating the operand.
794 + */
795 +static void
796 +reginsert(struct match_globals *g, char op, char* opnd)
797 +{
798 + register char *src;
799 + register char *dst;
800 + register char *place;
801 +
802 + if (g->regcode == &g->regdummy) {
803 + g->regsize += 3;
804 + return;
805 + }
806 +
807 + src = g->regcode;
808 + g->regcode += 3;
809 + dst = g->regcode;
810 + while (src > opnd)
811 + *--dst = *--src;
812 +
813 + place = opnd; /* Op node, where operand used to be. */
814 + *place++ = op;
815 + *place++ = '\0';
816 + *place++ = '\0';
817 +}
818 +
819 +/*
820 + - regtail - set the next-pointer at the end of a node chain
821 + */
822 +static void
823 +regtail(struct match_globals *g, char *p, char *val)
824 +{
825 + register char *scan;
826 + register char *temp;
827 + register int offset;
828 +
829 + if (p == &g->regdummy)
830 + return;
831 +
832 + /* Find last node. */
833 + scan = p;
834 + for (;;) {
835 + temp = regnext(g, scan);
836 + if (temp == NULL)
837 + break;
838 + scan = temp;
839 + }
840 +
841 + if (OP(scan) == BACK)
842 + offset = scan - val;
843 + else
844 + offset = val - scan;
845 + *(scan+1) = (offset>>8)&0377;
846 + *(scan+2) = offset&0377;
847 +}
848 +
849 +/*
850 + - regoptail - regtail on operand of first argument; nop if operandless
851 + */
852 +static void
853 +regoptail(struct match_globals *g, char *p, char *val)
854 +{
855 + /* "Operandless" and "op != BRANCH" are synonymous in practice. */
856 + if (p == NULL || p == &g->regdummy || OP(p) != BRANCH)
857 + return;
858 + regtail(g, OPERAND(p), val);
859 +}
860 +
861 +/*
862 + * regexec and friends
863 + */
864 +
865 +
866 +/*
867 + * Forwards.
868 + */
869 +STATIC int regtry(struct match_globals *g, regexp *prog, char *string);
870 +STATIC int regmatch(struct match_globals *g, char *prog);
871 +STATIC int regrepeat(struct match_globals *g, char *p);
872 +
873 +#ifdef DEBUG
874 +int regnarrate = 0;
875 +void regdump();
876 +STATIC char *regprop(char *op);
877 +#endif
878 +
879 +/*
880 + - regexec - match a regexp against a string
881 + */
882 +int
883 +regexec(regexp *prog, char *string)
884 +{
885 + register char *s;
886 + struct match_globals g;
887 +
888 + /* Be paranoid... */
889 + if (prog == NULL || string == NULL) {
890 + printk("<3>Regexp: NULL parameter\n");
891 + return(0);
892 + }
893 +
894 + /* Check validity of program. */
895 + if (UCHARAT(prog->program) != MAGIC) {
896 + printk("<3>Regexp: corrupted program\n");
897 + return(0);
898 + }
899 +
900 + /* If there is a "must appear" string, look for it. */
901 + if (prog->regmust != NULL) {
902 + s = string;
903 + while ((s = strchr(s, prog->regmust[0])) != NULL) {
904 + if (strncmp(s, prog->regmust, prog->regmlen) == 0)
905 + break; /* Found it. */
906 + s++;
907 + }
908 + if (s == NULL) /* Not present. */
909 + return(0);
910 + }
911 +
912 + /* Mark beginning of line for ^ . */
913 + g.regbol = string;
914 +
915 + /* Simplest case: anchored match need be tried only once. */
916 + if (prog->reganch)
917 + return(regtry(&g, prog, string));
918 +
919 + /* Messy cases: unanchored match. */
920 + s = string;
921 + if (prog->regstart != '\0')
922 + /* We know what char it must start with. */
923 + while ((s = strchr(s, prog->regstart)) != NULL) {
924 + if (regtry(&g, prog, s))
925 + return(1);
926 + s++;
927 + }
928 + else
929 + /* We don't -- general case. */
930 + do {
931 + if (regtry(&g, prog, s))
932 + return(1);
933 + } while (*s++ != '\0');
934 +
935 + /* Failure. */
936 + return(0);
937 +}
938 +
939 +/*
940 + - regtry - try match at specific point
941 + */
942 +static int /* 0 failure, 1 success */
943 +regtry(struct match_globals *g, regexp *prog, char *string)
944 +{
945 + register int i;
946 + register char **sp;
947 + register char **ep;
948 +
949 + g->reginput = string;
950 + g->regstartp = prog->startp;
951 + g->regendp = prog->endp;
952 +
953 + sp = prog->startp;
954 + ep = prog->endp;
955 + for (i = NSUBEXP; i > 0; i--) {
956 + *sp++ = NULL;
957 + *ep++ = NULL;
958 + }
959 + if (regmatch(g, prog->program + 1)) {
960 + prog->startp[0] = string;
961 + prog->endp[0] = g->reginput;
962 + return(1);
963 + } else
964 + return(0);
965 +}
966 +
967 +/*
968 + - regmatch - main matching routine
969 + *
970 + * Conceptually the strategy is simple: check to see whether the current
971 + * node matches, call self recursively to see whether the rest matches,
972 + * and then act accordingly. In practice we make some effort to avoid
973 + * recursion, in particular by going through "ordinary" nodes (that don't
974 + * need to know whether the rest of the match failed) by a loop instead of
975 + * by recursion.
976 + */
977 +static int /* 0 failure, 1 success */
978 +regmatch(struct match_globals *g, char *prog)
979 +{
980 + register char *scan = prog; /* Current node. */
981 + char *next; /* Next node. */
982 +
983 +#ifdef DEBUG
984 + if (scan != NULL && regnarrate)
985 + fprintf(stderr, "%s(\n", regprop(scan));
986 +#endif
987 + while (scan != NULL) {
988 +#ifdef DEBUG
989 + if (regnarrate)
990 + fprintf(stderr, "%s...\n", regprop(scan));
991 +#endif
992 + next = regnext(g, scan);
993 +
994 + switch (OP(scan)) {
995 + case BOL:
996 + if (g->reginput != g->regbol)
997 + return(0);
998 + break;
999 + case EOL:
1000 + if (*g->reginput != '\0')
1001 + return(0);
1002 + break;
1003 + case ANY:
1004 + if (*g->reginput == '\0')
1005 + return(0);
1006 + g->reginput++;
1007 + break;
1008 + case EXACTLY: {
1009 + register int len;
1010 + register char *opnd;
1011 +
1012 + opnd = OPERAND(scan);
1013 + /* Inline the first character, for speed. */
1014 + if (*opnd != *g->reginput)
1015 + return(0);
1016 + len = strlen(opnd);
1017 + if (len > 1 && strncmp(opnd, g->reginput, len) != 0)
1018 + return(0);
1019 + g->reginput += len;
1020 + }
1021 + break;
1022 + case ANYOF:
1023 + if (*g->reginput == '\0' || strchr(OPERAND(scan), *g->reginput) == NULL)
1024 + return(0);
1025 + g->reginput++;
1026 + break;
1027 + case ANYBUT:
1028 + if (*g->reginput == '\0' || strchr(OPERAND(scan), *g->reginput) != NULL)
1029 + return(0);
1030 + g->reginput++;
1031 + break;
1032 + case NOTHING:
1033 + case BACK:
1034 + break;
1035 + case OPEN+1:
1036 + case OPEN+2:
1037 + case OPEN+3:
1038 + case OPEN+4:
1039 + case OPEN+5:
1040 + case OPEN+6:
1041 + case OPEN+7:
1042 + case OPEN+8:
1043 + case OPEN+9: {
1044 + register int no;
1045 + register char *save;
1046 +
1047 + no = OP(scan) - OPEN;
1048 + save = g->reginput;
1049 +
1050 + if (regmatch(g, next)) {
1051 + /*
1052 + * Don't set startp if some later
1053 + * invocation of the same parentheses
1054 + * already has.
1055 + */
1056 + if (g->regstartp[no] == NULL)
1057 + g->regstartp[no] = save;
1058 + return(1);
1059 + } else
1060 + return(0);
1061 + }
1062 + break;
1063 + case CLOSE+1:
1064 + case CLOSE+2:
1065 + case CLOSE+3:
1066 + case CLOSE+4:
1067 + case CLOSE+5:
1068 + case CLOSE+6:
1069 + case CLOSE+7:
1070 + case CLOSE+8:
1071 + case CLOSE+9:
1072 + {
1073 + register int no;
1074 + register char *save;
1075 +
1076 + no = OP(scan) - CLOSE;
1077 + save = g->reginput;
1078 +
1079 + if (regmatch(g, next)) {
1080 + /*
1081 + * Don't set endp if some later
1082 + * invocation of the same parentheses
1083 + * already has.
1084 + */
1085 + if (g->regendp[no] == NULL)
1086 + g->regendp[no] = save;
1087 + return(1);
1088 + } else
1089 + return(0);
1090 + }
1091 + break;
1092 + case BRANCH: {
1093 + register char *save;
1094 +
1095 + if (OP(next) != BRANCH) /* No choice. */
1096 + next = OPERAND(scan); /* Avoid recursion. */
1097 + else {
1098 + do {
1099 + save = g->reginput;
1100 + if (regmatch(g, OPERAND(scan)))
1101 + return(1);
1102 + g->reginput = save;
1103 + scan = regnext(g, scan);
1104 + } while (scan != NULL && OP(scan) == BRANCH);
1105 + return(0);
1106 + /* NOTREACHED */
1107 + }
1108 + }
1109 + break;
1110 + case STAR:
1111 + case PLUS: {
1112 + register char nextch;
1113 + register int no;
1114 + register char *save;
1115 + register int min;
1116 +
1117 + /*
1118 + * Lookahead to avoid useless match attempts
1119 + * when we know what character comes next.
1120 + */
1121 + nextch = '\0';
1122 + if (OP(next) == EXACTLY)
1123 + nextch = *OPERAND(next);
1124 + min = (OP(scan) == STAR) ? 0 : 1;
1125 + save = g->reginput;
1126 + no = regrepeat(g, OPERAND(scan));
1127 + while (no >= min) {
1128 + /* If it could work, try it. */
1129 + if (nextch == '\0' || *g->reginput == nextch)
1130 + if (regmatch(g, next))
1131 + return(1);
1132 + /* Couldn't or didn't -- back up. */
1133 + no--;
1134 + g->reginput = save + no;
1135 + }
1136 + return(0);
1137 + }
1138 + break;
1139 + case END:
1140 + return(1); /* Success! */
1141 + break;
1142 + default:
1143 + printk("<3>Regexp: memory corruption\n");
1144 + return(0);
1145 + break;
1146 + }
1147 +
1148 + scan = next;
1149 + }
1150 +
1151 + /*
1152 + * We get here only if there's trouble -- normally "case END" is
1153 + * the terminating point.
1154 + */
1155 + printk("<3>Regexp: corrupted pointers\n");
1156 + return(0);
1157 +}
1158 +
1159 +/*
1160 + - regrepeat - repeatedly match something simple, report how many
1161 + */
1162 +static int
1163 +regrepeat(struct match_globals *g, char *p)
1164 +{
1165 + register int count = 0;
1166 + register char *scan;
1167 + register char *opnd;
1168 +
1169 + scan = g->reginput;
1170 + opnd = OPERAND(p);
1171 + switch (OP(p)) {
1172 + case ANY:
1173 + count = strlen(scan);
1174 + scan += count;
1175 + break;
1176 + case EXACTLY:
1177 + while (*opnd == *scan) {
1178 + count++;
1179 + scan++;
1180 + }
1181 + break;
1182 + case ANYOF:
1183 + while (*scan != '\0' && strchr(opnd, *scan) != NULL) {
1184 + count++;
1185 + scan++;
1186 + }
1187 + break;
1188 + case ANYBUT:
1189 + while (*scan != '\0' && strchr(opnd, *scan) == NULL) {
1190 + count++;
1191 + scan++;
1192 + }
1193 + break;
1194 + default: /* Oh dear. Called inappropriately. */
1195 + printk("<3>Regexp: internal foulup\n");
1196 + count = 0; /* Best compromise. */
1197 + break;
1198 + }
1199 + g->reginput = scan;
1200 +
1201 + return(count);
1202 +}
1203 +
1204 +/*
1205 + - regnext - dig the "next" pointer out of a node
1206 + */
1207 +static char*
1208 +regnext(struct match_globals *g, char *p)
1209 +{
1210 + register int offset;
1211 +
1212 + if (p == &g->regdummy)
1213 + return(NULL);
1214 +
1215 + offset = NEXT(p);
1216 + if (offset == 0)
1217 + return(NULL);
1218 +
1219 + if (OP(p) == BACK)
1220 + return(p-offset);
1221 + else
1222 + return(p+offset);
1223 +}
1224 +
1225 +#ifdef DEBUG
1226 +
1227 +STATIC char *regprop();
1228 +
1229 +/*
1230 + - regdump - dump a regexp onto stdout in vaguely comprehensible form
1231 + */
1232 +void
1233 +regdump(regexp *r)
1234 +{
1235 + register char *s;
1236 + register char op = EXACTLY; /* Arbitrary non-END op. */
1237 + register char *next;
1238 + /* extern char *strchr(); */
1239 +
1240 +
1241 + s = r->program + 1;
1242 + while (op != END) { /* While that wasn't END last time... */
1243 + op = OP(s);
1244 + printf("%2d%s", s-r->program, regprop(s)); /* Where, what. */
1245 + next = regnext(s);
1246 + if (next == NULL) /* Next ptr. */
1247 + printf("(0)");
1248 + else
1249 + printf("(%d)", (s-r->program)+(next-s));
1250 + s += 3;
1251 + if (op == ANYOF || op == ANYBUT || op == EXACTLY) {
1252 + /* Literal string, where present. */
1253 + while (*s != '\0') {
1254 + putchar(*s);
1255 + s++;
1256 + }
1257 + s++;
1258 + }
1259 + putchar('\n');
1260 + }
1261 +
1262 + /* Header fields of interest. */
1263 + if (r->regstart != '\0')
1264 + printf("start `%c' ", r->regstart);
1265 + if (r->reganch)
1266 + printf("anchored ");
1267 + if (r->regmust != NULL)
1268 + printf("must have \"%s\"", r->regmust);
1269 + printf("\n");
1270 +}
1271 +
1272 +/*
1273 + - regprop - printable representation of opcode
1274 + */
1275 +static char *
1276 +regprop(char *op)
1277 +{
1278 +#define BUFLEN 50
1279 + register char *p;
1280 + static char buf[BUFLEN];
1281 +
1282 + strcpy(buf, ":");
1283 +
1284 + switch (OP(op)) {
1285 + case BOL:
1286 + p = "BOL";
1287 + break;
1288 + case EOL:
1289 + p = "EOL";
1290 + break;
1291 + case ANY:
1292 + p = "ANY";
1293 + break;
1294 + case ANYOF:
1295 + p = "ANYOF";
1296 + break;
1297 + case ANYBUT:
1298 + p = "ANYBUT";
1299 + break;
1300 + case BRANCH:
1301 + p = "BRANCH";
1302 + break;
1303 + case EXACTLY:
1304 + p = "EXACTLY";
1305 + break;
1306 + case NOTHING:
1307 + p = "NOTHING";
1308 + break;
1309 + case BACK:
1310 + p = "BACK";
1311 + break;
1312 + case END:
1313 + p = "END";
1314 + break;
1315 + case OPEN+1:
1316 + case OPEN+2:
1317 + case OPEN+3:
1318 + case OPEN+4:
1319 + case OPEN+5:
1320 + case OPEN+6:
1321 + case OPEN+7:
1322 + case OPEN+8:
1323 + case OPEN+9:
1324 + snprintf(buf+strlen(buf),BUFLEN-strlen(buf), "OPEN%d", OP(op)-OPEN);
1325 + p = NULL;
1326 + break;
1327 + case CLOSE+1:
1328 + case CLOSE+2:
1329 + case CLOSE+3:
1330 + case CLOSE+4:
1331 + case CLOSE+5:
1332 + case CLOSE+6:
1333 + case CLOSE+7:
1334 + case CLOSE+8:
1335 + case CLOSE+9:
1336 + snprintf(buf+strlen(buf),BUFLEN-strlen(buf), "CLOSE%d", OP(op)-CLOSE);
1337 + p = NULL;
1338 + break;
1339 + case STAR:
1340 + p = "STAR";
1341 + break;
1342 + case PLUS:
1343 + p = "PLUS";
1344 + break;
1345 + default:
1346 + printk("<3>Regexp: corrupted opcode\n");
1347 + break;
1348 + }
1349 + if (p != NULL)
1350 + strncat(buf, p, BUFLEN-strlen(buf));
1351 + return(buf);
1352 +}
1353 +#endif
1354 +
1355 +
1356 diff --git a/net/netfilter/regexp/regexp.h b/net/netfilter/regexp/regexp.h
1357 new file mode 100644
1358 index 0000000..a72eba7
1359 --- /dev/null
1360 +++ b/net/netfilter/regexp/regexp.h
1361 @@ -0,0 +1,41 @@
1362 +/*
1363 + * Definitions etc. for regexp(3) routines.
1364 + *
1365 + * Caveat: this is V8 regexp(3) [actually, a reimplementation thereof],
1366 + * not the System V one.
1367 + */
1368 +
1369 +#ifndef REGEXP_H
1370 +#define REGEXP_H
1371 +
1372 +
1373 +/*
1374 +http://www.opensource.apple.com/darwinsource/10.3/expect-1/expect/expect.h ,
1375 +which contains a version of this library, says:
1376 +
1377 + *
1378 + * NSUBEXP must be at least 10, and no greater than 117 or the parser
1379 + * will not work properly.
1380 + *
1381 +
1382 +However, it looks rather like this library is limited to 10. If you think
1383 +otherwise, let us know.
1384 +*/
1385 +
1386 +#define NSUBEXP 10
1387 +typedef struct regexp {
1388 + char *startp[NSUBEXP];
1389 + char *endp[NSUBEXP];
1390 + char regstart; /* Internal use only. */
1391 + char reganch; /* Internal use only. */
1392 + char *regmust; /* Internal use only. */
1393 + int regmlen; /* Internal use only. */
1394 + char program[1]; /* Unwarranted chumminess with compiler. */
1395 +} regexp;
1396 +
1397 +regexp * regcomp(char *exp, int *patternsize);
1398 +int regexec(regexp *prog, char *string);
1399 +void regsub(regexp *prog, char *source, char *dest);
1400 +void regerror(char *s);
1401 +
1402 +#endif
1403 diff --git a/net/netfilter/regexp/regmagic.h b/net/netfilter/regexp/regmagic.h
1404 new file mode 100644
1405 index 0000000..5acf447
1406 --- /dev/null
1407 +++ b/net/netfilter/regexp/regmagic.h
1408 @@ -0,0 +1,5 @@
1409 +/*
1410 + * The first byte of the regexp internal "program" is actually this magic
1411 + * number; the start node begins in the second byte.
1412 + */
1413 +#define MAGIC 0234
1414 diff --git a/net/netfilter/regexp/regsub.c b/net/netfilter/regexp/regsub.c
1415 new file mode 100644
1416 index 0000000..339631f
1417 --- /dev/null
1418 +++ b/net/netfilter/regexp/regsub.c
1419 @@ -0,0 +1,95 @@
1420 +/*
1421 + * regsub
1422 + * @(#)regsub.c 1.3 of 2 April 86
1423 + *
1424 + * Copyright (c) 1986 by University of Toronto.
1425 + * Written by Henry Spencer. Not derived from licensed software.
1426 + *
1427 + * Permission is granted to anyone to use this software for any
1428 + * purpose on any computer system, and to redistribute it freely,
1429 + * subject to the following restrictions:
1430 + *
1431 + * 1. The author is not responsible for the consequences of use of
1432 + * this software, no matter how awful, even if they arise
1433 + * from defects in it.
1434 + *
1435 + * 2. The origin of this software must not be misrepresented, either
1436 + * by explicit claim or by omission.
1437 + *
1438 + * 3. Altered versions must be plainly marked as such, and must not
1439 + * be misrepresented as being the original software.
1440 + *
1441 + *
1442 + * This code was modified by Ethan Sommer to work within the kernel
1443 + * (it now uses kmalloc etc..)
1444 + *
1445 + */
1446 +#include "regexp.h"
1447 +#include "regmagic.h"
1448 +#include <linux/string.h>
1449 +
1450 +
1451 +#ifndef CHARBITS
1452 +#define UCHARAT(p) ((int)*(unsigned char *)(p))
1453 +#else
1454 +#define UCHARAT(p) ((int)*(p)&CHARBITS)
1455 +#endif
1456 +
1457 +#if 0
1458 +//void regerror(char * s)
1459 +//{
1460 +// printk("regexp(3): %s", s);
1461 +// /* NOTREACHED */
1462 +//}
1463 +#endif
1464 +
1465 +/*
1466 + - regsub - perform substitutions after a regexp match
1467 + */
1468 +void
1469 +regsub(regexp * prog, char * source, char * dest)
1470 +{
1471 + register char *src;
1472 + register char *dst;
1473 + register char c;
1474 + register int no;
1475 + register int len;
1476 +
1477 + /* Not necessary and gcc doesn't like it -MLS */
1478 + /*extern char *strncpy();*/
1479 +
1480 + if (prog == NULL || source == NULL || dest == NULL) {
1481 + regerror("NULL parm to regsub");
1482 + return;
1483 + }
1484 + if (UCHARAT(prog->program) != MAGIC) {
1485 + regerror("damaged regexp fed to regsub");
1486 + return;
1487 + }
1488 +
1489 + src = source;
1490 + dst = dest;
1491 + while ((c = *src++) != '\0') {
1492 + if (c == '&')
1493 + no = 0;
1494 + else if (c == '\\' && '0' <= *src && *src <= '9')
1495 + no = *src++ - '0';
1496 + else
1497 + no = -1;
1498 +
1499 + if (no < 0) { /* Ordinary character. */
1500 + if (c == '\\' && (*src == '\\' || *src == '&'))
1501 + c = *src++;
1502 + *dst++ = c;
1503 + } else if (prog->startp[no] != NULL && prog->endp[no] != NULL) {
1504 + len = prog->endp[no] - prog->startp[no];
1505 + (void) strncpy(dst, prog->startp[no], len);
1506 + dst += len;
1507 + if (len != 0 && *(dst-1) == '\0') { /* strncpy hit NUL. */
1508 + regerror("damaged match string");
1509 + return;
1510 + }
1511 + }
1512 + }
1513 + *dst++ = '\0';
1514 +}
1515 diff --git a/net/netfilter/xt_layer7.c b/net/netfilter/xt_layer7.c
1516 new file mode 100644
1517 index 0000000..51bb747
1518 --- /dev/null
1519 +++ b/net/netfilter/xt_layer7.c
1520 @@ -0,0 +1,625 @@
1521 +/*
1522 + Kernel module to match application layer (OSI layer 7) data in connections.
1523 +
1524 + http://l7-filter.sf.net
1525 +
1526 + (C) 2003-2009 Matthew Strait and Ethan Sommer.
1527 +
1528 + This program is free software; you can redistribute it and/or
1529 + modify it under the terms of the GNU General Public License
1530 + as published by the Free Software Foundation; either version
1531 + 2 of the License, or (at your option) any later version.
1532 + http://www.gnu.org/licenses/gpl.txt
1533 +
1534 + Based on ipt_string.c (C) 2000 Emmanuel Roger <winfield@freegates.be>,
1535 + xt_helper.c (C) 2002 Harald Welte and cls_layer7.c (C) 2003 Matthew Strait,
1536 + Ethan Sommer, Justin Levandoski.
1537 +*/
1538 +
1539 +#include <linux/spinlock.h>
1540 +#include <linux/version.h>
1541 +#include <net/ip.h>
1542 +#include <net/tcp.h>
1543 +#include <linux/module.h>
1544 +#include <linux/seq_file.h>
1545 +#include <linux/skbuff.h>
1546 +#include <linux/netfilter.h>
1547 +#include <net/netfilter/nf_conntrack.h>
1548 +#include <net/netfilter/nf_conntrack_core.h>
1549 +#include <net/netfilter/nf_conntrack_extend.h>
1550 +#include <net/netfilter/nf_conntrack_acct.h>
1551 +#include <linux/netfilter/x_tables.h>
1552 +#include <linux/netfilter/xt_layer7.h>
1553 +#include <linux/ctype.h>
1554 +#include <linux/proc_fs.h>
1555 +
1556 +#include "regexp/regexp.c"
1557 +
1558 +MODULE_LICENSE("GPL");
1559 +MODULE_AUTHOR("Matthew Strait <quadong@users.sf.net>, Ethan Sommer <sommere@users.sf.net>");
1560 +MODULE_DESCRIPTION("iptables application layer match module");
1561 +MODULE_ALIAS("ipt_layer7");
1562 +MODULE_VERSION("2.21");
1563 +
1564 +static int maxdatalen = 2048; // this is the default
1565 +module_param(maxdatalen, int, 0444);
1566 +MODULE_PARM_DESC(maxdatalen, "maximum bytes of data looked at by l7-filter");
1567 +#ifdef CONFIG_NETFILTER_XT_MATCH_LAYER7_DEBUG
1568 + #define DPRINTK(format,args...) printk(format,##args)
1569 +#else
1570 + #define DPRINTK(format,args...)
1571 +#endif
1572 +
1573 +/* Number of packets whose data we look at.
1574 +This can be modified through /proc/net/layer7_numpackets */
1575 +static int num_packets = 10;
1576 +
1577 +static struct pattern_cache {
1578 + char * regex_string;
1579 + regexp * pattern;
1580 + struct pattern_cache * next;
1581 +} * first_pattern_cache = NULL;
1582 +
1583 +DEFINE_SPINLOCK(l7_lock);
1584 +
1585 +static int total_acct_packets(struct nf_conn *ct)
1586 +{
1587 + struct nf_conn_counter *acct;
1588 +
1589 + BUG_ON(ct == NULL);
1590 + acct = nf_conn_acct_find(ct);
1591 + if (!acct)
1592 + return 0;
1593 + return (atomic64_read(&acct[IP_CT_DIR_ORIGINAL].packets) + atomic64_read(&acct[IP_CT_DIR_REPLY].packets));
1594 +}
1595 +
1596 +#ifdef CONFIG_IP_NF_MATCH_LAYER7_DEBUG
1597 +/* Converts an unfriendly string into a friendly one by
1598 +replacing unprintables with periods and all whitespace with " ". */
1599 +static char * friendly_print(unsigned char * s)
1600 +{
1601 + char * f = kmalloc(strlen(s) + 1, GFP_ATOMIC);
1602 + int i;
1603 +
1604 + if(!f) {
1605 + if (net_ratelimit())
1606 + printk(KERN_ERR "layer7: out of memory in "
1607 + "friendly_print, bailing.\n");
1608 + return NULL;
1609 + }
1610 +
1611 + for(i = 0; i < strlen(s); i++){
1612 + if(isprint(s[i]) && s[i] < 128) f[i] = s[i];
1613 + else if(isspace(s[i])) f[i] = ' ';
1614 + else f[i] = '.';
1615 + }
1616 + f[i] = '\0';
1617 + return f;
1618 +}
1619 +
1620 +static char dec2hex(int i)
1621 +{
1622 + switch (i) {
1623 + case 0 ... 9:
1624 + return (i + '0');
1625 + break;
1626 + case 10 ... 15:
1627 + return (i - 10 + 'a');
1628 + break;
1629 + default:
1630 + if (net_ratelimit())
1631 + printk("layer7: Problem in dec2hex\n");
1632 + return '\0';
1633 + }
1634 +}
1635 +
1636 +static char * hex_print(unsigned char * s)
1637 +{
1638 + char * g = kmalloc(strlen(s)*3 + 1, GFP_ATOMIC);
1639 + int i;
1640 +
1641 + if(!g) {
1642 + if (net_ratelimit())
1643 + printk(KERN_ERR "layer7: out of memory in hex_print, "
1644 + "bailing.\n");
1645 + return NULL;
1646 + }
1647 +
1648 + for(i = 0; i < strlen(s); i++) {
1649 + g[i*3 ] = dec2hex(s[i]/16);
1650 + g[i*3 + 1] = dec2hex(s[i]%16);
1651 + g[i*3 + 2] = ' ';
1652 + }
1653 + g[i*3] = '\0';
1654 +
1655 + return g;
1656 +}
1657 +#endif // DEBUG
1658 +
1659 +/* Use instead of regcomp. As we expect to be seeing the same regexps over and
1660 +over again, it make sense to cache the results. */
1661 +static regexp * compile_and_cache(const char * regex_string,
1662 + const char * protocol)
1663 +{
1664 + struct pattern_cache * node = first_pattern_cache;
1665 + struct pattern_cache * last_pattern_cache = first_pattern_cache;
1666 + struct pattern_cache * tmp;
1667 + unsigned int len;
1668 +
1669 + while (node != NULL) {
1670 + if (!strcmp(node->regex_string, regex_string))
1671 + return node->pattern;
1672 +
1673 + last_pattern_cache = node;/* points at the last non-NULL node */
1674 + node = node->next;
1675 + }
1676 +
1677 + /* If we reach the end of the list, then we have not yet cached
1678 + the pattern for this regex. Let's do that now.
1679 + Be paranoid about running out of memory to avoid list corruption. */
1680 + tmp = kmalloc(sizeof(struct pattern_cache), GFP_ATOMIC);
1681 +
1682 + if(!tmp) {
1683 + if (net_ratelimit())
1684 + printk(KERN_ERR "layer7: out of memory in "
1685 + "compile_and_cache, bailing.\n");
1686 + return NULL;
1687 + }
1688 +
1689 + tmp->regex_string = kmalloc(strlen(regex_string) + 1, GFP_ATOMIC);
1690 + tmp->pattern = kmalloc(sizeof(struct regexp), GFP_ATOMIC);
1691 + tmp->next = NULL;
1692 +
1693 + if(!tmp->regex_string || !tmp->pattern) {
1694 + if (net_ratelimit())
1695 + printk(KERN_ERR "layer7: out of memory in "
1696 + "compile_and_cache, bailing.\n");
1697 + kfree(tmp->regex_string);
1698 + kfree(tmp->pattern);
1699 + kfree(tmp);
1700 + return NULL;
1701 + }
1702 +
1703 + /* Ok. The new node is all ready now. */
1704 + node = tmp;
1705 +
1706 + if(first_pattern_cache == NULL) /* list is empty */
1707 + first_pattern_cache = node; /* make node the beginning */
1708 + else
1709 + last_pattern_cache->next = node; /* attach node to the end */
1710 +
1711 + /* copy the string and compile the regex */
1712 + len = strlen(regex_string);
1713 + DPRINTK("About to compile this: \"%s\"\n", regex_string);
1714 + node->pattern = regcomp((char *)regex_string, &len);
1715 + if ( !node->pattern ) {
1716 + if (net_ratelimit())
1717 + printk(KERN_ERR "layer7: Error compiling regexp "
1718 + "\"%s\" (%s)\n",
1719 + regex_string, protocol);
1720 + /* pattern is now cached as NULL, so we won't try again. */
1721 + }
1722 +
1723 + strcpy(node->regex_string, regex_string);
1724 + return node->pattern;
1725 +}
1726 +
1727 +static int can_handle(const struct sk_buff *skb)
1728 +{
1729 + if(!ip_hdr(skb)) /* not IP */
1730 + return 0;
1731 + if(ip_hdr(skb)->protocol != IPPROTO_TCP &&
1732 + ip_hdr(skb)->protocol != IPPROTO_UDP &&
1733 + ip_hdr(skb)->protocol != IPPROTO_ICMP)
1734 + return 0;
1735 + return 1;
1736 +}
1737 +
1738 +/* Returns offset the into the skb->data that the application data starts */
1739 +static int app_data_offset(const struct sk_buff *skb)
1740 +{
1741 + /* In case we are ported somewhere (ebtables?) where ip_hdr(skb)
1742 + isn't set, this can be gotten from 4*(skb->data[0] & 0x0f) as well. */
1743 + int ip_hl = 4*ip_hdr(skb)->ihl;
1744 +
1745 + if( ip_hdr(skb)->protocol == IPPROTO_TCP ) {
1746 + /* 12 == offset into TCP header for the header length field.
1747 + Can't get this with skb->h.th->doff because the tcphdr
1748 + struct doesn't get set when routing (this is confirmed to be
1749 + true in Netfilter as well as QoS.) */
1750 + int tcp_hl = 4*(skb->data[ip_hl + 12] >> 4);
1751 +
1752 + return ip_hl + tcp_hl;
1753 + } else if( ip_hdr(skb)->protocol == IPPROTO_UDP ) {
1754 + return ip_hl + 8; /* UDP header is always 8 bytes */
1755 + } else if( ip_hdr(skb)->protocol == IPPROTO_ICMP ) {
1756 + return ip_hl + 8; /* ICMP header is 8 bytes */
1757 + } else {
1758 + if (net_ratelimit())
1759 + printk(KERN_ERR "layer7: tried to handle unknown "
1760 + "protocol!\n");
1761 + return ip_hl + 8; /* something reasonable */
1762 + }
1763 +}
1764 +
1765 +/* handles whether there's a match when we aren't appending data anymore */
1766 +static int match_no_append(struct nf_conn * conntrack,
1767 + struct nf_conn * master_conntrack,
1768 + enum ip_conntrack_info ctinfo,
1769 + enum ip_conntrack_info master_ctinfo,
1770 + const struct xt_layer7_info * info)
1771 +{
1772 + /* If we're in here, throw the app data away */
1773 + if(master_conntrack->layer7.app_data != NULL) {
1774 +
1775 + #ifdef CONFIG_IP_NF_MATCH_LAYER7_DEBUG
1776 + if(!master_conntrack->layer7.app_proto) {
1777 + char * f =
1778 + friendly_print(master_conntrack->layer7.app_data);
1779 + char * g =
1780 + hex_print(master_conntrack->layer7.app_data);
1781 + DPRINTK("\nl7-filter gave up after %d bytes "
1782 + "(%d packets):\n%s\n",
1783 + strlen(f), total_acct_packets(master_conntrack), f);
1784 + kfree(f);
1785 + DPRINTK("In hex: %s\n", g);
1786 + kfree(g);
1787 + }
1788 + #endif
1789 +
1790 + kfree(master_conntrack->layer7.app_data);
1791 + master_conntrack->layer7.app_data = NULL; /* don't free again */
1792 + }
1793 +
1794 + if(master_conntrack->layer7.app_proto){
1795 + /* Here child connections set their .app_proto (for /proc) */
1796 + if(!conntrack->layer7.app_proto) {
1797 + conntrack->layer7.app_proto =
1798 + kmalloc(strlen(master_conntrack->layer7.app_proto)+1,
1799 + GFP_ATOMIC);
1800 + if(!conntrack->layer7.app_proto){
1801 + if (net_ratelimit())
1802 + printk(KERN_ERR "layer7: out of memory "
1803 + "in match_no_append, "
1804 + "bailing.\n");
1805 + return 1;
1806 + }
1807 + strcpy(conntrack->layer7.app_proto,
1808 + master_conntrack->layer7.app_proto);
1809 + }
1810 +
1811 + return (!strcmp(master_conntrack->layer7.app_proto,
1812 + info->protocol));
1813 + }
1814 + else {
1815 + /* If not classified, set to "unknown" to distinguish from
1816 + connections that are still being tested. */
1817 + master_conntrack->layer7.app_proto =
1818 + kmalloc(strlen("unknown")+1, GFP_ATOMIC);
1819 + if(!master_conntrack->layer7.app_proto){
1820 + if (net_ratelimit())
1821 + printk(KERN_ERR "layer7: out of memory in "
1822 + "match_no_append, bailing.\n");
1823 + return 1;
1824 + }
1825 + strcpy(master_conntrack->layer7.app_proto, "unknown");
1826 + return 0;
1827 + }
1828 +}
1829 +
1830 +/* add the new app data to the conntrack. Return number of bytes added. */
1831 +static int add_datastr(char *target, int offset, char *app_data, int len)
1832 +{
1833 + int length = 0, i;
1834 + if (!target) return 0;
1835 +
1836 + /* Strip nulls. Make everything lower case (our regex lib doesn't
1837 + do case insensitivity). Add it to the end of the current data. */
1838 + for(i = 0; i < maxdatalen-offset-1 && i < len; i++) {
1839 + if(app_data[i] != '\0') {
1840 + /* the kernel version of tolower mungs 'upper ascii' */
1841 + target[length+offset] =
1842 + isascii(app_data[i])?
1843 + tolower(app_data[i]) : app_data[i];
1844 + length++;
1845 + }
1846 + }
1847 + target[length+offset] = '\0';
1848 +
1849 + return length;
1850 +}
1851 +
1852 +/* add the new app data to the conntrack. Return number of bytes added. */
1853 +static int add_data(struct nf_conn * master_conntrack,
1854 + char * app_data, int appdatalen)
1855 +{
1856 + int length;
1857 +
1858 + length = add_datastr(master_conntrack->layer7.app_data, master_conntrack->layer7.app_data_len, app_data, appdatalen);
1859 + master_conntrack->layer7.app_data_len += length;
1860 +
1861 + return length;
1862 +}
1863 +
1864 +/* taken from drivers/video/modedb.c */
1865 +static int my_atoi(const char *s)
1866 +{
1867 + int val = 0;
1868 +
1869 + for (;; s++) {
1870 + switch (*s) {
1871 + case '0'...'9':
1872 + val = 10*val+(*s-'0');
1873 + break;
1874 + default:
1875 + return val;
1876 + }
1877 + }
1878 +}
1879 +
1880 +static int layer7_numpackets_proc_show(struct seq_file *s, void *p) {
1881 + seq_printf(s, "%d\n", num_packets);
1882 +
1883 + return 0;
1884 +}
1885 +
1886 +static int layer7_numpackets_proc_open(struct inode *inode, struct file *file) {
1887 + return single_open(file, layer7_numpackets_proc_show, NULL);
1888 +}
1889 +
1890 +/* Read in num_packets from userland */
1891 +static ssize_t layer7_numpackets_write_proc(struct file* file, const char __user *buffer,
1892 + size_t count, loff_t *data) {
1893 + char value[1024];
1894 + int new_num_packets;
1895 +
1896 + if (copy_from_user(&value, buffer, sizeof(value)))
1897 + return -EFAULT;
1898 +
1899 + new_num_packets = my_atoi(value);
1900 +
1901 + if ((new_num_packets < 1) || (new_num_packets > 99)) {
1902 + printk(KERN_WARNING "layer7: numpackets must be between 1 and 99\n");
1903 + return -EFAULT;
1904 + }
1905 +
1906 + num_packets = new_num_packets;
1907 +
1908 + return count;
1909 +}
1910 +
1911 +static bool match(const struct sk_buff *skbin, struct xt_action_param *par)
1912 +{
1913 + /* sidestep const without getting a compiler warning... */
1914 + struct sk_buff * skb = (struct sk_buff *)skbin;
1915 +
1916 + const struct xt_layer7_info * info = par->matchinfo;
1917 +
1918 + enum ip_conntrack_info master_ctinfo, ctinfo;
1919 + struct nf_conn *master_conntrack, *conntrack;
1920 + unsigned char *app_data, *tmp_data;
1921 + unsigned int pattern_result, appdatalen;
1922 + regexp * comppattern;
1923 +
1924 + /* Be paranoid/incompetent - lock the entire match function. */
1925 + spin_lock_bh(&l7_lock);
1926 +
1927 + if(!can_handle(skb)){
1928 + DPRINTK("layer7: This is some protocol I can't handle.\n");
1929 + spin_unlock_bh(&l7_lock);
1930 + return info->invert;
1931 + }
1932 +
1933 + /* Treat parent & all its children together as one connection, except
1934 + for the purpose of setting conntrack->layer7.app_proto in the actual
1935 + connection. This makes /proc/net/ip_conntrack more satisfying. */
1936 + if(!(conntrack = nf_ct_get(skb, &ctinfo)) ||
1937 + !(master_conntrack=nf_ct_get(skb,&master_ctinfo))){
1938 + DPRINTK("layer7: couldn't get conntrack.\n");
1939 + spin_unlock_bh(&l7_lock);
1940 + return info->invert;
1941 + }
1942 +
1943 + /* Try to get a master conntrack (and its master etc) for FTP, etc. */
1944 + while (master_ct(master_conntrack) != NULL)
1945 + master_conntrack = master_ct(master_conntrack);
1946 +
1947 + /* if we've classified it or seen too many packets */
1948 + if(!info->pkt && (total_acct_packets(master_conntrack) > num_packets ||
1949 + master_conntrack->layer7.app_proto)) {
1950 +
1951 + pattern_result = match_no_append(conntrack, master_conntrack,
1952 + ctinfo, master_ctinfo, info);
1953 +
1954 + /* skb->cb[0] == seen. Don't do things twice if there are
1955 + multiple l7 rules. I'm not sure that using cb for this purpose
1956 + is correct, even though it says "put your private variables
1957 + there". But it doesn't look like it is being used for anything
1958 + else in the skbs that make it here. */
1959 + skb->cb[0] = 1; /* marking it seen here's probably irrelevant */
1960 +
1961 + spin_unlock_bh(&l7_lock);
1962 + return (pattern_result ^ info->invert);
1963 + }
1964 +
1965 + if(skb_is_nonlinear(skb)){
1966 + if(skb_linearize(skb) != 0){
1967 + if (net_ratelimit())
1968 + printk(KERN_ERR "layer7: failed to linearize "
1969 + "packet, bailing.\n");
1970 + spin_unlock_bh(&l7_lock);
1971 + return info->invert;
1972 + }
1973 + }
1974 +
1975 + /* now that the skb is linearized, it's safe to set these. */
1976 + app_data = skb->data + app_data_offset(skb);
1977 + appdatalen = skb_tail_pointer(skb) - app_data;
1978 +
1979 + /* the return value gets checked later, when we're ready to use it */
1980 + comppattern = compile_and_cache(info->pattern, info->protocol);
1981 +
1982 + if (info->pkt) {
1983 + tmp_data = kmalloc(maxdatalen, GFP_ATOMIC);
1984 + if(!tmp_data){
1985 + if (net_ratelimit())
1986 + printk(KERN_ERR "layer7: out of memory in match, bailing.\n");
1987 + return info->invert;
1988 + }
1989 +
1990 + tmp_data[0] = '\0';
1991 + add_datastr(tmp_data, 0, app_data, appdatalen);
1992 + pattern_result = ((comppattern && regexec(comppattern, tmp_data)) ? 1 : 0);
1993 +
1994 + kfree(tmp_data);
1995 + tmp_data = NULL;
1996 + spin_unlock_bh(&l7_lock);
1997 +
1998 + return (pattern_result ^ info->invert);
1999 + }
2000 +
2001 + /* On the first packet of a connection, allocate space for app data */
2002 + if(total_acct_packets(master_conntrack) == 1 && !skb->cb[0] &&
2003 + !master_conntrack->layer7.app_data){
2004 + master_conntrack->layer7.app_data =
2005 + kmalloc(maxdatalen, GFP_ATOMIC);
2006 + if(!master_conntrack->layer7.app_data){
2007 + if (net_ratelimit())
2008 + printk(KERN_ERR "layer7: out of memory in "
2009 + "match, bailing.\n");
2010 + spin_unlock_bh(&l7_lock);
2011 + return info->invert;
2012 + }
2013 +
2014 + master_conntrack->layer7.app_data[0] = '\0';
2015 + }
2016 +
2017 + /* Can be here, but unallocated, if numpackets is increased near
2018 + the beginning of a connection */
2019 + if(master_conntrack->layer7.app_data == NULL){
2020 + spin_unlock_bh(&l7_lock);
2021 + return info->invert; /* unmatched */
2022 + }
2023 +
2024 + if(!skb->cb[0]){
2025 + int newbytes;
2026 + newbytes = add_data(master_conntrack, app_data, appdatalen);
2027 +
2028 + if(newbytes == 0) { /* didn't add any data */
2029 + skb->cb[0] = 1;
2030 + /* Didn't match before, not going to match now */
2031 + spin_unlock_bh(&l7_lock);
2032 + return info->invert;
2033 + }
2034 + }
2035 +
2036 + /* If looking for "unknown", then never match. "Unknown" means that
2037 + we've given up; we're still trying with these packets. */
2038 + if(!strcmp(info->protocol, "unknown")) {
2039 + pattern_result = 0;
2040 + /* If looking for "unset", then always match. "Unset" means that we
2041 + haven't yet classified the connection. */
2042 + } else if(!strcmp(info->protocol, "unset")) {
2043 + pattern_result = 2;
2044 + DPRINTK("layer7: matched unset: not yet classified "
2045 + "(%d/%d packets)\n",
2046 + total_acct_packets(master_conntrack), num_packets);
2047 + /* If the regexp failed to compile, don't bother running it */
2048 + } else if(comppattern &&
2049 + regexec(comppattern, master_conntrack->layer7.app_data)){
2050 + DPRINTK("layer7: matched %s\n", info->protocol);
2051 + pattern_result = 1;
2052 + } else pattern_result = 0;
2053 +
2054 + if(pattern_result == 1) {
2055 + master_conntrack->layer7.app_proto =
2056 + kmalloc(strlen(info->protocol)+1, GFP_ATOMIC);
2057 + if(!master_conntrack->layer7.app_proto){
2058 + if (net_ratelimit())
2059 + printk(KERN_ERR "layer7: out of memory in "
2060 + "match, bailing.\n");
2061 + spin_unlock_bh(&l7_lock);
2062 + return (pattern_result ^ info->invert);
2063 + }
2064 + strcpy(master_conntrack->layer7.app_proto, info->protocol);
2065 + } else if(pattern_result > 1) { /* cleanup from "unset" */
2066 + pattern_result = 1;
2067 + }
2068 +
2069 + /* mark the packet seen */
2070 + skb->cb[0] = 1;
2071 +
2072 + spin_unlock_bh(&l7_lock);
2073 + return (pattern_result ^ info->invert);
2074 +}
2075 +
2076 +// load nf_conntrack_ipv4
2077 +static int check(const struct xt_mtchk_param *par)
2078 +{
2079 + if (nf_ct_l3proto_try_module_get(par->match->family) < 0) {
2080 + printk(KERN_WARNING "can't load conntrack support for "
2081 + "proto=%d\n", par->match->family);
2082 + return -EINVAL;
2083 + }
2084 + return 0;
2085 +}
2086 +
2087 +
2088 +static void destroy(const struct xt_mtdtor_param *par)
2089 +{
2090 + nf_ct_l3proto_module_put(par->match->family);
2091 +}
2092 +
2093 +static struct xt_match xt_layer7_match[] __read_mostly = {
2094 +{
2095 + .name = "layer7",
2096 + .family = AF_INET,
2097 + .checkentry = check,
2098 + .match = match,
2099 + .destroy = destroy,
2100 + .matchsize = sizeof(struct xt_layer7_info),
2101 + .me = THIS_MODULE
2102 +}
2103 +};
2104 +
2105 +static const struct file_operations layer7_numpackets_proc_fops = {
2106 + .owner = THIS_MODULE,
2107 + .open = layer7_numpackets_proc_open,
2108 + .read = seq_read,
2109 + .llseek = seq_lseek,
2110 + .release = single_release,
2111 + .write = layer7_numpackets_write_proc,
2112 +};
2113 +
2114 +static int __init xt_layer7_init(void)
2115 +{
2116 + need_conntrack();
2117 +
2118 + // Register proc interface
2119 + proc_create_data("layer7_numpackets", 0644,
2120 + init_net.proc_net, &layer7_numpackets_proc_fops, NULL);
2121 +
2122 + if(maxdatalen < 1) {
2123 + printk(KERN_WARNING "layer7: maxdatalen can't be < 1, "
2124 + "using 1\n");
2125 + maxdatalen = 1;
2126 + }
2127 + /* This is not a hard limit. It's just here to prevent people from
2128 + bringing their slow machines to a grinding halt. */
2129 + else if(maxdatalen > 65536) {
2130 + printk(KERN_WARNING "layer7: maxdatalen can't be > 65536, "
2131 + "using 65536\n");
2132 + maxdatalen = 65536;
2133 + }
2134 + return xt_register_matches(xt_layer7_match,
2135 + ARRAY_SIZE(xt_layer7_match));
2136 +}
2137 +
2138 +static void __exit xt_layer7_fini(void)
2139 +{
2140 + remove_proc_entry("layer7_numpackets", init_net.proc_net);
2141 + xt_unregister_matches(xt_layer7_match, ARRAY_SIZE(xt_layer7_match));
2142 +}
2143 +
2144 +module_init(xt_layer7_init);
2145 +module_exit(xt_layer7_fini);