1/* $NetBSD: ipsec_mbuf.c,v 1.12 2011/05/16 10:05:23 drochner Exp $ */
2/*-
3 * Copyright (c) 2002, 2003 Sam Leffler, Errno Consulting
4 * All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 *
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25 * SUCH DAMAGE.
26 *
27 * $FreeBSD: /repoman/r/ncvs/src/sys/netipsec/ipsec_mbuf.c,v 1.5.2.2 2003/03/28 20:32:53 sam Exp $
28 */
29
30#include <sys/cdefs.h>
31__KERNEL_RCSID(0, "$NetBSD: ipsec_mbuf.c,v 1.12 2011/05/16 10:05:23 drochner Exp $");
32
33/*
34 * IPsec-specific mbuf routines.
35 */
36
37#ifdef __FreeBSD__
38#include "opt_param.h"
39#endif
40
41#include <sys/param.h>
42#include <sys/systm.h>
43#include <sys/mbuf.h>
44#include <sys/socket.h>
45
46#include <net/route.h>
47#include <netinet/in.h>
48
49#include <netipsec/ipsec.h>
50#include <netipsec/ipsec_var.h>
51#include <netipsec/ipsec_private.h>
52
53#include <netipsec/ipsec_osdep.h>
54#include <net/net_osdep.h>
55
56/*
57 * Create a writable copy of the mbuf chain. While doing this
58 * we compact the chain with a goal of producing a chain with
59 * at most two mbufs. The second mbuf in this chain is likely
60 * to be a cluster. The primary purpose of this work is to create
61 * a writable packet for encryption, compression, etc. The
62 * secondary goal is to linearize the data so the data can be
63 * passed to crypto hardware in the most efficient manner possible.
64 */
65struct mbuf *
66m_clone(struct mbuf *m0)
67{
68 struct mbuf *m, *mprev;
69 struct mbuf *n, *mfirst, *mlast;
70 int len, off;
71
72 IPSEC_ASSERT(m0 != NULL, ("m_clone: null mbuf"));
73
74 mprev = NULL;
75 for (m = m0; m != NULL; m = mprev->m_next) {
76 /*
77 * Regular mbufs are ignored unless there's a cluster
78 * in front of it that we can use to coalesce. We do
79 * the latter mainly so later clusters can be coalesced
80 * also w/o having to handle them specially (i.e. convert
81 * mbuf+cluster -> cluster). This optimization is heavily
82 * influenced by the assumption that we're running over
83 * Ethernet where MCLBYTES is large enough that the max
84 * packet size will permit lots of coalescing into a
85 * single cluster. This in turn permits efficient
86 * crypto operations, especially when using hardware.
87 */
88 if ((m->m_flags & M_EXT) == 0) {
89 if (mprev && (mprev->m_flags & M_EXT) &&
90 m->m_len <= M_TRAILINGSPACE(mprev)) {
91 /* XXX: this ignores mbuf types */
92 memcpy(mtod(mprev, char *) + mprev->m_len,
93 mtod(m, char *), m->m_len);
94 mprev->m_len += m->m_len;
95 mprev->m_next = m->m_next; /* unlink from chain */
96 m_free(m); /* reclaim mbuf */
97 IPSEC_STATINC(IPSEC_STAT_MBCOALESCED);
98 } else {
99 mprev = m;
100 }
101 continue;
102 }
103 /*
104 * Writable mbufs are left alone (for now). Note
105 * that for 4.x systems it's not possible to identify
106 * whether or not mbufs with external buffers are
107 * writable unless they use clusters.
108 */
109 if (M_EXT_WRITABLE(m)) {
110 mprev = m;
111 continue;
112 }
113
114 /*
115 * Not writable, replace with a copy or coalesce with
116 * the previous mbuf if possible (since we have to copy
117 * it anyway, we try to reduce the number of mbufs and
118 * clusters so that future work is easier).
119 */
120 IPSEC_ASSERT(m->m_flags & M_EXT,
121 ("m_clone: m_flags 0x%x", m->m_flags));
122 /* NB: we only coalesce into a cluster or larger */
123 if (mprev != NULL && (mprev->m_flags & M_EXT) &&
124 m->m_len <= M_TRAILINGSPACE(mprev)) {
125 /* XXX: this ignores mbuf types */
126 memcpy(mtod(mprev, char *) + mprev->m_len,
127 mtod(m, char *), m->m_len);
128 mprev->m_len += m->m_len;
129 mprev->m_next = m->m_next; /* unlink from chain */
130 m_free(m); /* reclaim mbuf */
131 IPSEC_STATINC(IPSEC_STAT_CLCOALESCED);
132 continue;
133 }
134
135 /*
136 * Allocate new space to hold the copy...
137 */
138 /* XXX why can M_PKTHDR be set past the first mbuf? */
139 if (mprev == NULL && (m->m_flags & M_PKTHDR)) {
140 /*
141 * NB: if a packet header is present we must
142 * allocate the mbuf separately from any cluster
143 * because M_MOVE_PKTHDR will smash the data
144 * pointer and drop the M_EXT marker.
145 */
146 MGETHDR(n, M_DONTWAIT, m->m_type);
147 if (n == NULL) {
148 m_freem(m0);
149 return (NULL);
150 }
151 M_MOVE_PKTHDR(n, m);
152 MCLGET(n, M_DONTWAIT);
153 if ((n->m_flags & M_EXT) == 0) {
154 m_free(n);
155 m_freem(m0);
156 return (NULL);
157 }
158 } else {
159 n = m_getcl(M_DONTWAIT, m->m_type, m->m_flags);
160 if (n == NULL) {
161 m_freem(m0);
162 return (NULL);
163 }
164 }
165 /*
166 * ... and copy the data. We deal with jumbo mbufs
167 * (i.e. m_len > MCLBYTES) by splitting them into
168 * clusters. We could just malloc a buffer and make
169 * it external but too many device drivers don't know
170 * how to break up the non-contiguous memory when
171 * doing DMA.
172 */
173 len = m->m_len;
174 off = 0;
175 mfirst = n;
176 mlast = NULL;
177 for (;;) {
178 int cc = min(len, MCLBYTES);
179 memcpy(mtod(n, char *), mtod(m, char *) + off, cc);
180 n->m_len = cc;
181 if (mlast != NULL)
182 mlast->m_next = n;
183 mlast = n;
184 IPSEC_STATINC(IPSEC_STAT_CLCOPIED);
185
186 len -= cc;
187 if (len <= 0)
188 break;
189 off += cc;
190
191 n = m_getcl(M_DONTWAIT, m->m_type, m->m_flags);
192 if (n == NULL) {
193 m_freem(mfirst);
194 m_freem(m0);
195 return (NULL);
196 }
197 }
198 n->m_next = m->m_next;
199 if (mprev == NULL)
200 m0 = mfirst; /* new head of chain */
201 else
202 mprev->m_next = mfirst; /* replace old mbuf */
203 m_free(m); /* release old mbuf */
204 mprev = mfirst;
205 }
206 return (m0);
207}
208
209/*
210 * Make space for a new header of length hlen at skip bytes
211 * into the packet. When doing this we allocate new mbufs only
212 * when absolutely necessary. The mbuf where the new header
213 * is to go is returned together with an offset into the mbuf.
214 * If NULL is returned then the mbuf chain may have been modified;
215 * the caller is assumed to always free the chain.
216 */
217struct mbuf *
218m_makespace(struct mbuf *m0, int skip, int hlen, int *off)
219{
220 struct mbuf *m;
221 unsigned remain;
222
223 IPSEC_ASSERT(m0 != NULL, ("m_dmakespace: null mbuf"));
224 IPSEC_ASSERT(hlen < MHLEN, ("m_makespace: hlen too big: %u", hlen));
225
226 for (m = m0; m && skip > m->m_len; m = m->m_next)
227 skip -= m->m_len;
228 if (m == NULL)
229 return (NULL);
230 /*
231 * At this point skip is the offset into the mbuf m
232 * where the new header should be placed. Figure out
233 * if there's space to insert the new header. If so,
234 * and copying the remainder makese sense then do so.
235 * Otherwise insert a new mbuf in the chain, splitting
236 * the contents of m as needed.
237 */
238 remain = m->m_len - skip; /* data to move */
239 if (hlen > M_TRAILINGSPACE(m)) {
240 struct mbuf *n0, *n, **np;
241 int todo, len, done, alloc;
242
243 n0 = NULL;
244 np = &n0;
245 alloc = 0;
246 done = 0;
247 todo = remain;
248 while (todo > 0) {
249 if (todo > MHLEN) {
250 n = m_getcl(M_DONTWAIT, m->m_type, 0);
251 len = MCLBYTES;
252 }
253 else {
254 n = m_get(M_DONTWAIT, m->m_type);
255 len = MHLEN;
256 }
257 if (n == NULL) {
258 m_freem(n0);
259 return NULL;
260 }
261 *np = n;
262 np = &n->m_next;
263 alloc++;
264 len = min(todo, len);
265 memcpy(n->m_data, mtod(m, char *) + skip + done, len);
266 n->m_len = len;
267 done += len;
268 todo -= len;
269 }
270
271 if (hlen <= M_TRAILINGSPACE(m) + remain) {
272 m->m_len = skip + hlen;
273 *off = skip;
274 if (n0 != NULL) {
275 *np = m->m_next;
276 m->m_next = n0;
277 }
278 }
279 else {
280 n = m_get(M_DONTWAIT, m->m_type);
281 if (n == NULL) {
282 m_freem(n0);
283 return NULL;
284 }
285 alloc++;
286
287 if ((n->m_next = n0) == NULL)
288 np = &n->m_next;
289 n0 = n;
290
291 *np = m->m_next;
292 m->m_next = n0;
293
294 n->m_len = hlen;
295 m->m_len = skip;
296
297 m = n; /* header is at front ... */
298 *off = 0; /* ... of new mbuf */
299 }
300
301 IPSEC_STATADD(IPSEC_STAT_MBINSERTED, alloc);
302 } else {
303 /*
304 * Copy the remainder to the back of the mbuf
305 * so there's space to write the new header.
306 */
307 /* XXX can this be memcpy? does it handle overlap? */
308 ovbcopy(mtod(m, char *) + skip,
309 mtod(m, char *) + skip + hlen, remain);
310 m->m_len += hlen;
311 *off = skip;
312 }
313 m0->m_pkthdr.len += hlen; /* adjust packet length */
314 return m;
315}
316
317/*
318 * m_pad(m, n) pads <m> with <n> bytes at the end. The packet header
319 * length is updated, and a pointer to the first byte of the padding
320 * (which is guaranteed to be all in one mbuf) is returned.
321 */
322void *
323m_pad(struct mbuf *m, int n)
324{
325 register struct mbuf *m0, *m1;
326 register int len, pad;
327 void *retval;
328
329 if (n <= 0) { /* No stupid arguments. */
330 DPRINTF(("m_pad: pad length invalid (%d)\n", n));
331 m_freem(m);
332 return NULL;
333 }
334
335 len = m->m_pkthdr.len;
336 pad = n;
337 m0 = m;
338
339 while (m0->m_len < len) {
340IPSEC_ASSERT(m0->m_next != NULL, ("m_pad: m0 null, len %u m_len %u", len, m0->m_len));/*XXX*/
341 len -= m0->m_len;
342 m0 = m0->m_next;
343 }
344
345 if (m0->m_len != len) {
346 DPRINTF(("m_pad: length mismatch (should be %d instead of %d)\n",
347 m->m_pkthdr.len, m->m_pkthdr.len + m0->m_len - len));
348
349 m_freem(m);
350 return NULL;
351 }
352
353 /* Check for zero-length trailing mbufs, and find the last one. */
354 for (m1 = m0; m1->m_next; m1 = m1->m_next) {
355 if (m1->m_next->m_len != 0) {
356 DPRINTF(("m_pad: length mismatch (should be %d "
357 "instead of %d)\n",
358 m->m_pkthdr.len,
359 m->m_pkthdr.len + m1->m_next->m_len));
360
361 m_freem(m);
362 return NULL;
363 }
364
365 m0 = m1->m_next;
366 }
367
368 if (pad > M_TRAILINGSPACE(m0)) {
369 /* Add an mbuf to the chain. */
370 MGET(m1, M_DONTWAIT, MT_DATA);
371 if (m1 == 0) {
372 m_freem(m0);
373 DPRINTF(("m_pad: unable to get extra mbuf\n"));
374 return NULL;
375 }
376
377 m0->m_next = m1;
378 m0 = m1;
379 m0->m_len = 0;
380 }
381
382 retval = m0->m_data + m0->m_len;
383 m0->m_len += pad;
384 m->m_pkthdr.len += pad;
385
386 return retval;
387}
388
389/*
390 * Remove hlen data at offset skip in the packet. This is used by
391 * the protocols strip protocol headers and associated data (e.g. IV,
392 * authenticator) on input.
393 */
394int
395m_striphdr(struct mbuf *m, int skip, int hlen)
396{
397 struct mbuf *m1;
398 int roff;
399
400 /* Find beginning of header */
401 m1 = m_getptr(m, skip, &roff);
402 if (m1 == NULL)
403 return (EINVAL);
404
405 /* Remove the header and associated data from the mbuf. */
406 if (roff == 0) {
407 /* The header was at the beginning of the mbuf */
408 IPSEC_STATINC(IPSEC_STAT_INPUT_FRONT);
409 m_adj(m1, hlen);
410 if ((m1->m_flags & M_PKTHDR) == 0)
411 m->m_pkthdr.len -= hlen;
412 } else if (roff + hlen >= m1->m_len) {
413 struct mbuf *mo;
414
415 /*
416 * Part or all of the header is at the end of this mbuf,
417 * so first let's remove the remainder of the header from
418 * the beginning of the remainder of the mbuf chain, if any.
419 */
420 IPSEC_STATINC(IPSEC_STAT_INPUT_END);
421 if (roff + hlen > m1->m_len) {
422 /* Adjust the next mbuf by the remainder */
423 m_adj(m1->m_next, roff + hlen - m1->m_len);
424
425 /* The second mbuf is guaranteed not to have a pkthdr... */
426 m->m_pkthdr.len -= (roff + hlen - m1->m_len);
427 }
428
429 /* Now, let's unlink the mbuf chain for a second...*/
430 mo = m1->m_next;
431 m1->m_next = NULL;
432
433 /* ...and trim the end of the first part of the chain...sick */
434 m_adj(m1, -(m1->m_len - roff));
435 if ((m1->m_flags & M_PKTHDR) == 0)
436 m->m_pkthdr.len -= (m1->m_len - roff);
437
438 /* Finally, let's relink */
439 m1->m_next = mo;
440 } else {
441 /*
442 * The header lies in the "middle" of the mbuf; copy
443 * the remainder of the mbuf down over the header.
444 */
445 IPSEC_STATINC(IPSEC_STAT_INPUT_MIDDLE);
446 ovbcopy(mtod(m1, u_char *) + roff + hlen,
447 mtod(m1, u_char *) + roff,
448 m1->m_len - (roff + hlen));
449 m1->m_len -= hlen;
450 m->m_pkthdr.len -= hlen;
451 }
452 return (0);
453}
454
455/*
456 * Diagnostic routine to check mbuf alignment as required by the
457 * crypto device drivers (that use DMA).
458 */
459void
460m_checkalignment(const char* where, struct mbuf *m0, int off, int len)
461{
462 int roff;
463 struct mbuf *m = m_getptr(m0, off, &roff);
464 void *addr;
465
466 if (m == NULL)
467 return;
468 printf("%s (off %u len %u): ", where, off, len);
469 addr = mtod(m, char *) + roff;
470 do {
471 int mlen;
472
473 if (((uintptr_t) addr) & 3) {
474 printf("addr misaligned %p,", addr);
475 break;
476 }
477 mlen = m->m_len;
478 if (mlen > len)
479 mlen = len;
480 len -= mlen;
481 if (len && (mlen & 3)) {
482 printf("len mismatch %u,", mlen);
483 break;
484 }
485 m = m->m_next;
486 addr = m ? mtod(m, void *) : NULL;
487 } while (m && len > 0);
488 for (m = m0; m; m = m->m_next)
489 printf(" [%p:%u]", mtod(m, void *), m->m_len);
490 printf("\n");
491}
492