2 * Copyright (c) 2002, 2003 Sam Leffler, Errno Consulting
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * $FreeBSD: src/sys/netipsec/ipsec_mbuf.c,v 1.5.2.2 2003/03/28 20:32:53 sam Exp $
27 * $DragonFly: src/sys/netproto/ipsec/ipsec_mbuf.c,v 1.6 2004/10/15 22:59:10 hsu Exp $
31 * IPsec-specific mbuf routines.
34 #include "opt_param.h"
36 #include <sys/param.h>
37 #include <sys/systm.h>
39 #include <sys/socket.h>
41 #include <net/route.h>
42 #include <netinet/in.h>
44 #include <netproto/ipsec/ipsec.h>
46 extern struct mbuf *m_getptr(struct mbuf *, int, int *);
49 * Create a writable copy of the mbuf chain. While doing this
50 * we compact the chain with a goal of producing a chain with
51 * at most two mbufs. The second mbuf in this chain is likely
52 * to be a cluster. The primary purpose of this work is to create
53 * a writable packet for encryption, compression, etc. The
54 * secondary goal is to linearize the data so the data can be
55 * passed to crypto hardware in the most efficient manner possible.
58 m_clone(struct mbuf *m0)
60 struct mbuf *m, *mprev;
61 struct mbuf *n, *mfirst, *mlast;
64 KASSERT(m0 != NULL, ("m_clone: null mbuf"));
67 for (m = m0; m != NULL; m = mprev->m_next) {
69 * Regular mbufs are ignored unless there's a cluster
70 * in front of it that we can use to coalesce. We do
71 * the latter mainly so later clusters can be coalesced
72 * also w/o having to handle them specially (i.e. convert
73 * mbuf+cluster -> cluster). This optimization is heavily
74 * influenced by the assumption that we're running over
75 * Ethernet where MCLBYTES is large enough that the max
76 * packet size will permit lots of coalescing into a
77 * single cluster. This in turn permits efficient
78 * crypto operations, especially when using hardware.
80 if ((m->m_flags & M_EXT) == 0) {
81 if (mprev && (mprev->m_flags & M_EXT) &&
82 m->m_len <= M_TRAILINGSPACE(mprev)) {
83 /* XXX: this ignores mbuf types */
84 memcpy(mtod(mprev, caddr_t) + mprev->m_len,
85 mtod(m, caddr_t), m->m_len);
86 mprev->m_len += m->m_len;
87 mprev->m_next = m->m_next; /* unlink from chain */
88 m_free(m); /* reclaim mbuf */
89 newipsecstat.ips_mbcoalesced++;
96 * Writable mbufs are left alone (for now). Note
97 * that for 4.x systems it's not possible to identify
98 * whether or not mbufs with external buffers are
99 * writable unless they use clusters.
101 if (M_EXT_WRITABLE(m)) {
107 * Not writable, replace with a copy or coalesce with
108 * the previous mbuf if possible (since we have to copy
109 * it anyway, we try to reduce the number of mbufs and
110 * clusters so that future work is easier).
112 KASSERT(m->m_flags & M_EXT,
113 ("m_clone: m_flags 0x%x", m->m_flags));
114 /* NB: we only coalesce into a cluster or larger */
115 if (mprev != NULL && (mprev->m_flags & M_EXT) &&
116 m->m_len <= M_TRAILINGSPACE(mprev)) {
117 /* XXX: this ignores mbuf types */
118 memcpy(mtod(mprev, caddr_t) + mprev->m_len,
119 mtod(m, caddr_t), m->m_len);
120 mprev->m_len += m->m_len;
121 mprev->m_next = m->m_next; /* unlink from chain */
122 m_free(m); /* reclaim mbuf */
123 newipsecstat.ips_clcoalesced++;
128 * Allocate new space to hold the copy...
130 /* XXX why can M_PKTHDR be set past the first mbuf? */
131 if (mprev == NULL && (m->m_flags & M_PKTHDR)) {
133 * NB: if a packet header is present we must
134 * allocate the mbuf separately from any cluster
135 * because M_MOVE_PKTHDR will smash the data
136 * pointer and drop the M_EXT marker.
138 MGETHDR(n, MB_DONTWAIT, m->m_type);
144 MCLGET(n, MB_DONTWAIT);
145 if ((n->m_flags & M_EXT) == 0) {
151 n = m_getcl(MB_DONTWAIT, m->m_type, m->m_flags);
158 * ... and copy the data. We deal with jumbo mbufs
159 * (i.e. m_len > MCLBYTES) by splitting them into
160 * clusters. We could just malloc a buffer and make
161 * it external but too many device drivers don't know
162 * how to break up the non-contiguous memory when
170 int cc = min(len, MCLBYTES);
171 memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + off, cc);
176 newipsecstat.ips_clcopied++;
183 n = m_getcl(MB_DONTWAIT, m->m_type, m->m_flags);
190 n->m_next = m->m_next;
192 m0 = mfirst; /* new head of chain */
194 mprev->m_next = mfirst; /* replace old mbuf */
195 m_free(m); /* release old mbuf */
202 * Make space for a new header of length hlen at skip bytes
203 * into the packet. When doing this we allocate new mbufs only
204 * when absolutely necessary. The mbuf where the new header
205 * is to go is returned together with an offset into the mbuf.
206 * If NULL is returned then the mbuf chain may have been modified;
207 * the caller is assumed to always free the chain.
210 m_makespace(struct mbuf *m0, int skip, int hlen, int *off)
215 KASSERT(m0 != NULL, ("m_dmakespace: null mbuf"));
216 KASSERT(hlen < MHLEN, ("m_makespace: hlen too big: %u", hlen));
218 for (m = m0; m && skip > m->m_len; m = m->m_next)
223 * At this point skip is the offset into the mbuf m
224 * where the new header should be placed. Figure out
225 * if there's space to insert the new header. If so,
226 * and copying the remainder makese sense then do so.
227 * Otherwise insert a new mbuf in the chain, splitting
228 * the contents of m as needed.
230 remain = m->m_len - skip; /* data to move */
231 if (hlen > M_TRAILINGSPACE(m)) {
234 /* XXX code doesn't handle clusters XXX */
235 KASSERT(remain < MLEN,
236 ("m_makespace: remainder too big: %u", remain));
238 * Not enough space in m, split the contents
239 * of m, inserting new mbufs as required.
241 * NB: this ignores mbuf types.
243 MGET(n, MB_DONTWAIT, MT_DATA);
246 n->m_next = m->m_next; /* splice new mbuf */
248 newipsecstat.ips_mbinserted++;
249 if (hlen <= M_TRAILINGSPACE(m) + remain) {
251 * New header fits in the old mbuf if we copy
252 * the remainder; just do the copy to the new
253 * mbuf and we're good to go.
255 memcpy(mtod(n, caddr_t),
256 mtod(m, caddr_t) + skip, remain);
258 m->m_len = skip + hlen;
262 * No space in the old mbuf for the new header.
263 * Make space in the new mbuf and check the
264 * remainder'd data fits too. If not then we
265 * must allocate an additional mbuf (yech).
268 if (remain + hlen > M_TRAILINGSPACE(n)) {
271 MGET(n2, MB_DONTWAIT, MT_DATA);
272 /* NB: new mbuf is on chain, let caller free */
276 memcpy(mtod(n2, caddr_t),
277 mtod(m, caddr_t) + skip, remain);
279 /* splice in second mbuf */
280 n2->m_next = n->m_next;
282 newipsecstat.ips_mbinserted++;
284 memcpy(mtod(n, caddr_t) + hlen,
285 mtod(m, caddr_t) + skip, remain);
290 m = n; /* header is at front ... */
291 *off = 0; /* ... of new mbuf */
295 * Copy the remainder to the back of the mbuf
296 * so there's space to write the new header.
298 /* XXX can this be memcpy? does it handle overlap? */
299 ovbcopy(mtod(m, caddr_t) + skip,
300 mtod(m, caddr_t) + skip + hlen, remain);
304 m0->m_pkthdr.len += hlen; /* adjust packet length */
309 * m_pad(m, n) pads <m> with <n> bytes at the end. The packet header
310 * length is updated, and a pointer to the first byte of the padding
311 * (which is guaranteed to be all in one mbuf) is returned.
314 m_pad(struct mbuf *m, int n)
316 struct mbuf *m0, *m1;
320 if (n <= 0) { /* No stupid arguments. */
321 DPRINTF(("m_pad: pad length invalid (%d)\n", n));
326 len = m->m_pkthdr.len;
330 while (m0->m_len < len) {
331 KASSERT(m0->m_next != NULL, ("m_pad: m0 null, len %u m_len %u", len, m0->m_len));/*XXX*/
336 if (m0->m_len != len) {
337 DPRINTF(("m_pad: length mismatch (should be %d instead of %d)\n",
338 m->m_pkthdr.len, m->m_pkthdr.len + m0->m_len - len));
344 /* Check for zero-length trailing mbufs, and find the last one. */
345 for (m1 = m0; m1->m_next; m1 = m1->m_next) {
346 if (m1->m_next->m_len != 0) {
347 DPRINTF(("m_pad: length mismatch (should be %d "
350 m->m_pkthdr.len + m1->m_next->m_len));
359 if (pad > M_TRAILINGSPACE(m0)) {
360 /* Add an mbuf to the chain. */
361 MGET(m1, MB_DONTWAIT, MT_DATA);
364 DPRINTF(("m_pad: unable to get extra mbuf\n"));
373 retval = m0->m_data + m0->m_len;
375 m->m_pkthdr.len += pad;
381 * Remove hlen data at offset skip in the packet. This is used by
382 * the protocols strip protocol headers and associated data (e.g. IV,
383 * authenticator) on input.
386 m_striphdr(struct mbuf *m, int skip, int hlen)
391 /* Find beginning of header */
392 m1 = m_getptr(m, skip, &roff);
396 /* Remove the header and associated data from the mbuf. */
398 /* The header was at the beginning of the mbuf */
399 newipsecstat.ips_input_front++;
401 if ((m1->m_flags & M_PKTHDR) == 0)
402 m->m_pkthdr.len -= hlen;
403 } else if (roff + hlen >= m1->m_len) {
407 * Part or all of the header is at the end of this mbuf,
408 * so first let's remove the remainder of the header from
409 * the beginning of the remainder of the mbuf chain, if any.
411 newipsecstat.ips_input_end++;
412 if (roff + hlen > m1->m_len) {
413 /* Adjust the next mbuf by the remainder */
414 m_adj(m1->m_next, roff + hlen - m1->m_len);
416 /* The second mbuf is guaranteed not to have a pkthdr... */
417 m->m_pkthdr.len -= (roff + hlen - m1->m_len);
420 /* Now, let's unlink the mbuf chain for a second...*/
424 /* ...and trim the end of the first part of the chain...sick */
425 m_adj(m1, -(m1->m_len - roff));
426 if ((m1->m_flags & M_PKTHDR) == 0)
427 m->m_pkthdr.len -= (m1->m_len - roff);
429 /* Finally, let's relink */
433 * The header lies in the "middle" of the mbuf; copy
434 * the remainder of the mbuf down over the header.
436 newipsecstat.ips_input_middle++;
437 bcopy(mtod(m1, u_char *) + roff + hlen,
438 mtod(m1, u_char *) + roff,
439 m1->m_len - (roff + hlen));
441 m->m_pkthdr.len -= hlen;
447 * Diagnostic routine to check mbuf alignment as required by the
448 * crypto device drivers (that use DMA).
451 m_checkalignment(const char* where, struct mbuf *m0, int off, int len)
454 struct mbuf *m = m_getptr(m0, off, &roff);
459 printf("%s (off %u len %u): ", where, off, len);
460 addr = mtod(m, caddr_t) + roff;
464 if (((uintptr_t) addr) & 3) {
465 printf("addr misaligned %p,", addr);
472 if (len && (mlen & 3)) {
473 printf("len mismatch %u,", mlen);
477 addr = m ? mtod(m, caddr_t) : NULL;
478 } while (m && len > 0);
479 for (m = m0; m; m = m->m_next)
480 printf(" [%p:%u]", mtod(m, caddr_t), m->m_len);