Home | History | Annotate | Line # | Download | only in netipsec
      1 /*	$NetBSD: ipsec_mbuf.c,v 1.30 2018/12/22 13:11:38 maxv Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 2002, 2003 Sam Leffler, Errno Consulting
      5  * All rights reserved.
      6  *
      7  * Redistribution and use in source and binary forms, with or without
      8  * modification, are permitted provided that the following conditions
      9  * are met:
     10  * 1. Redistributions of source code must retain the above copyright
     11  *    notice, this list of conditions and the following disclaimer.
     12  * 2. Redistributions in binary form must reproduce the above copyright
     13  *    notice, this list of conditions and the following disclaimer in the
     14  *    documentation and/or other materials provided with the distribution.
     15  *
     16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
     17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
     20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     26  * SUCH DAMAGE.
     27  *
     28  * $FreeBSD: sys/netipsec/ipsec_mbuf.c,v 1.5.2.2 2003/03/28 20:32:53 sam Exp $
     29  */
     30 
     31 #include <sys/cdefs.h>
     32 __KERNEL_RCSID(0, "$NetBSD: ipsec_mbuf.c,v 1.30 2018/12/22 13:11:38 maxv Exp $");
     33 
     34 /*
     35  * IPsec-specific mbuf routines.
     36  */
     37 
     38 #include <sys/param.h>
     39 #include <sys/systm.h>
     40 #include <sys/mbuf.h>
     41 
     42 #include <netipsec/ipsec.h>
     43 #include <netipsec/ipsec_var.h>
     44 #include <netipsec/ipsec_private.h>
     45 
     46 /*
     47  * Create a writable copy of the mbuf chain.  While doing this
     48  * we compact the chain with a goal of producing a chain with
     49  * at most two mbufs.  The second mbuf in this chain is likely
     50  * to be a cluster.  The primary purpose of this work is to create
     51  * a writable packet for encryption, compression, etc.  The
     52  * secondary goal is to linearize the data so the data can be
     53  * passed to crypto hardware in the most efficient manner possible.
     54  */
     55 struct mbuf *
     56 m_clone(struct mbuf *m0)
     57 {
     58 	struct mbuf *m, *mprev;
     59 	struct mbuf *n, *mfirst, *mlast;
     60 	int len, off;
     61 
     62 	KASSERT(m0 != NULL);
     63 
     64 	mprev = NULL;
     65 	for (m = m0; m != NULL; m = mprev->m_next) {
     66 		/*
     67 		 * Regular mbufs are ignored unless there's a cluster
     68 		 * in front of it that we can use to coalesce.
     69 		 */
     70 		if ((m->m_flags & M_EXT) == 0) {
     71 			if (mprev && (mprev->m_flags & M_EXT) &&
     72 			    m->m_len <= M_TRAILINGSPACE(mprev)) {
     73 				memcpy(mtod(mprev, char *) + mprev->m_len,
     74 				    mtod(m, char *), m->m_len);
     75 				mprev->m_len += m->m_len;
     76 				mprev->m_next = m_free(m);
     77 				IPSEC_STATINC(IPSEC_STAT_MBCOALESCED);
     78 			} else {
     79 				mprev = m;
     80 			}
     81 			continue;
     82 		}
     83 
     84 		/*
     85 		 * Writable mbufs are left alone.
     86 		 */
     87 		if (!M_READONLY(m)) {
     88 			mprev = m;
     89 			continue;
     90 		}
     91 
     92 		/*
     93 		 * Not writable, replace with a copy or coalesce with
     94 		 * the previous mbuf if possible (since we have to copy
     95 		 * it anyway, we try to reduce the number of mbufs and
     96 		 * clusters so that future work is easier).
     97 		 */
     98 
     99 		/* We only coalesce into a cluster. */
    100 		if (mprev != NULL && (mprev->m_flags & M_EXT) &&
    101 		    m->m_len <= M_TRAILINGSPACE(mprev)) {
    102 			memcpy(mtod(mprev, char *) + mprev->m_len,
    103 			    mtod(m, char *), m->m_len);
    104 			mprev->m_len += m->m_len;
    105 			mprev->m_next = m_free(m);
    106 			IPSEC_STATINC(IPSEC_STAT_CLCOALESCED);
    107 			continue;
    108 		}
    109 
    110 		/*
    111 		 * Allocate new space to hold the copy...
    112 		 */
    113 		if (mprev == NULL && (m->m_flags & M_PKTHDR)) {
    114 			MGETHDR(n, M_DONTWAIT, m->m_type);
    115 			if (n == NULL) {
    116 				m_freem(m0);
    117 				return NULL;
    118 			}
    119 			m_move_pkthdr(n, m);
    120 			MCLGET(n, M_DONTWAIT);
    121 			if ((n->m_flags & M_EXT) == 0) {
    122 				m_free(n);
    123 				m_freem(m0);
    124 				return NULL;
    125 			}
    126 		} else {
    127 			n = m_getcl(M_DONTWAIT, m->m_type, m->m_flags);
    128 			if (n == NULL) {
    129 				m_freem(m0);
    130 				return NULL;
    131 			}
    132 		}
    133 
    134 		/*
    135 		 * ... and copy the data.  We deal with jumbo mbufs
    136 		 * (i.e. m_len > MCLBYTES) by splitting them into
    137 		 * clusters.  We could just malloc a buffer and make
    138 		 * it external but too many device drivers don't know
    139 		 * how to break up the non-contiguous memory when
    140 		 * doing DMA.
    141 		 */
    142 		len = m->m_len;
    143 		off = 0;
    144 		mfirst = n;
    145 		mlast = NULL;
    146 		for (;;) {
    147 			const int cc = uimin(len, MCLBYTES);
    148 			memcpy(mtod(n, char *), mtod(m, char *) + off, cc);
    149 			n->m_len = cc;
    150 			if (mlast != NULL)
    151 				mlast->m_next = n;
    152 			mlast = n;
    153 			IPSEC_STATINC(IPSEC_STAT_CLCOPIED);
    154 
    155 			len -= cc;
    156 			if (len <= 0)
    157 				break;
    158 			off += cc;
    159 
    160 			n = m_getcl(M_DONTWAIT, m->m_type, m->m_flags);
    161 			if (n == NULL) {
    162 				m_freem(mfirst);
    163 				m_freem(m0);
    164 				return NULL;
    165 			}
    166 		}
    167 		n->m_next = m->m_next;
    168 		if (mprev == NULL)
    169 			m0 = mfirst;		/* new head of chain */
    170 		else
    171 			mprev->m_next = mfirst;	/* replace old mbuf */
    172 		m_free(m);			/* release old mbuf */
    173 		mprev = mfirst;
    174 	}
    175 
    176 	return m0;
    177 }
    178 
    179 /*
    180  * Make space for a new header of length hlen at skip bytes
    181  * into the packet.  When doing this we allocate new mbufs only
    182  * when absolutely necessary.  The mbuf where the new header
    183  * is to go is returned together with an offset into the mbuf.
    184  * If NULL is returned then the mbuf chain may have been modified;
    185  * the caller is assumed to always free the chain.
    186  */
    187 struct mbuf *
    188 m_makespace(struct mbuf *m0, int skip, int hlen, int *off)
    189 {
    190 	struct mbuf *m;
    191 	unsigned remain;
    192 
    193 	KASSERT(m0 != NULL);
    194 	KASSERT(m0->m_flags & M_PKTHDR);
    195 	KASSERTMSG(hlen < MHLEN, "hlen too big: %u", hlen);
    196 
    197 	for (m = m0; m && skip > m->m_len; m = m->m_next)
    198 		skip -= m->m_len;
    199 	if (m == NULL)
    200 		return NULL;
    201 
    202 	/*
    203 	 * At this point skip is the offset into the mbuf m
    204 	 * where the new header should be placed.  Figure out
    205 	 * if there's space to insert the new header.  If so,
    206 	 * and copying the remainder makes sense then do so.
    207 	 * Otherwise insert a new mbuf in the chain, splitting
    208 	 * the contents of m as needed.
    209 	 */
    210 	remain = m->m_len - skip;		/* data to move */
    211 	if (hlen > M_TRAILINGSPACE(m)) {
    212 		struct mbuf *n0, *n, **np;
    213 		int todo, len, done, alloc;
    214 
    215 		n0 = NULL;
    216 		np = &n0;
    217 		alloc = 0;
    218 		done = 0;
    219 		todo = remain;
    220 		while (todo > 0) {
    221 			if (todo > MHLEN) {
    222 				n = m_getcl(M_DONTWAIT, m->m_type, 0);
    223 				len = MCLBYTES;
    224 			} else {
    225 				n = m_get(M_DONTWAIT, m->m_type);
    226 				len = MHLEN;
    227 			}
    228 			if (n == NULL) {
    229 				m_freem(n0);
    230 				return NULL;
    231 			}
    232 			*np = n;
    233 			np = &n->m_next;
    234 			alloc++;
    235 			len = uimin(todo, len);
    236 			memcpy(n->m_data, mtod(m, char *) + skip + done, len);
    237 			n->m_len = len;
    238 			done += len;
    239 			todo -= len;
    240 		}
    241 
    242 		if (hlen <= M_TRAILINGSPACE(m) + remain) {
    243 			m->m_len = skip + hlen;
    244 			*off = skip;
    245 			if (n0 != NULL) {
    246 				*np = m->m_next;
    247 				m->m_next = n0;
    248 			}
    249 		} else {
    250 			n = m_get(M_DONTWAIT, m->m_type);
    251 			if (n == NULL) {
    252 				m_freem(n0);
    253 				return NULL;
    254 			}
    255 			alloc++;
    256 
    257 			if ((n->m_next = n0) == NULL)
    258 				np = &n->m_next;
    259 			n0 = n;
    260 
    261 			*np = m->m_next;
    262 			m->m_next = n0;
    263 
    264 			n->m_len = hlen;
    265 			m->m_len = skip;
    266 
    267 			m = n;			/* header is at front ... */
    268 			*off = 0;		/* ... of new mbuf */
    269 		}
    270 
    271 		IPSEC_STATADD(IPSEC_STAT_MBINSERTED, alloc);
    272 	} else {
    273 		/*
    274 		 * Copy the remainder to the back of the mbuf
    275 		 * so there's space to write the new header.
    276 		 */
    277 		/* XXX can this be memcpy? does it handle overlap? */
    278 		memmove(mtod(m, char *) + skip + hlen,
    279 			mtod(m, char *) + skip, remain);
    280 		m->m_len += hlen;
    281 		*off = skip;
    282 	}
    283 
    284 	m0->m_pkthdr.len += hlen;		/* adjust packet length */
    285 	return m;
    286 }
    287 
    288 /*
    289  * m_pad(m, n) pads <m> with <n> bytes at the end. The packet header
    290  * length is updated, and a pointer to the first byte of the padding
    291  * (which is guaranteed to be all in one mbuf) is returned.
    292  */
    293 void *
    294 m_pad(struct mbuf *m, int n)
    295 {
    296 	register struct mbuf *m0, *m1;
    297 	register int len, pad;
    298 	void *retval;
    299 
    300 	if (__predict_false(n > MLEN)) {
    301 		panic("%s: %d > MLEN", __func__, n);
    302 	}
    303 	KASSERT(m->m_flags & M_PKTHDR);
    304 
    305 	len = m->m_pkthdr.len;
    306 	pad = n;
    307 	m0 = m;
    308 
    309 	while (m0->m_len < len) {
    310 		KASSERTMSG(m0->m_next != NULL,
    311 		    "m0 null, len %u m_len %u", len, m0->m_len);
    312 		len -= m0->m_len;
    313 		m0 = m0->m_next;
    314 	}
    315 
    316 	if (m0->m_len != len) {
    317 		IPSECLOG(LOG_DEBUG,
    318 		    "length mismatch (should be %d instead of %d)\n",
    319 		    m->m_pkthdr.len, m->m_pkthdr.len + m0->m_len - len);
    320 		m_freem(m);
    321 		return NULL;
    322 	}
    323 
    324 	/* Check for zero-length trailing mbufs, and find the last one. */
    325 	for (m1 = m0; m1->m_next; m1 = m1->m_next) {
    326 		if (m1->m_next->m_len != 0) {
    327 			IPSECLOG(LOG_DEBUG,
    328 			    "length mismatch (should be %d instead of %d)\n",
    329 			    m->m_pkthdr.len,
    330 			    m->m_pkthdr.len + m1->m_next->m_len);
    331 			m_freem(m);
    332 			return NULL;
    333 		}
    334 
    335 		m0 = m1->m_next;
    336 	}
    337 
    338 	if (pad > M_TRAILINGSPACE(m0)) {
    339 		/* Add an mbuf to the chain. */
    340 		MGET(m1, M_DONTWAIT, MT_DATA);
    341 		if (m1 == NULL) {
    342 			m_freem(m);
    343 			IPSECLOG(LOG_DEBUG, "unable to get extra mbuf\n");
    344 			return NULL;
    345 		}
    346 
    347 		m0->m_next = m1;
    348 		m0 = m1;
    349 		m0->m_len = 0;
    350 	}
    351 
    352 	retval = m0->m_data + m0->m_len;
    353 	m0->m_len += pad;
    354 	m->m_pkthdr.len += pad;
    355 
    356 	return retval;
    357 }
    358 
    359 /*
    360  * Remove hlen data at offset skip in the packet.  This is used by
    361  * the protocols strip protocol headers and associated data (e.g. IV,
    362  * authenticator) on input.
    363  */
    364 int
    365 m_striphdr(struct mbuf *m, int skip, int hlen)
    366 {
    367 	struct mbuf *m1;
    368 	int roff;
    369 
    370 	KASSERT(m->m_flags & M_PKTHDR);
    371 
    372 	/* Find beginning of header */
    373 	m1 = m_getptr(m, skip, &roff);
    374 	if (m1 == NULL)
    375 		return EINVAL;
    376 
    377 	/* Remove the header and associated data from the mbuf. */
    378 	if (roff == 0) {
    379 		/* The header was at the beginning of the mbuf */
    380 		IPSEC_STATINC(IPSEC_STAT_INPUT_FRONT);
    381 		m_adj(m1, hlen);
    382 		if (m1 != m)
    383 			m->m_pkthdr.len -= hlen;
    384 	} else if (roff + hlen >= m1->m_len) {
    385 		struct mbuf *mo;
    386 		int adjlen;
    387 
    388 		/*
    389 		 * Part or all of the header is at the end of this mbuf,
    390 		 * so first let's remove the remainder of the header from
    391 		 * the beginning of the remainder of the mbuf chain, if any.
    392 		 */
    393 		IPSEC_STATINC(IPSEC_STAT_INPUT_END);
    394 		if (roff + hlen > m1->m_len) {
    395 			adjlen = roff + hlen - m1->m_len;
    396 
    397 			/* Adjust the next mbuf by the remainder */
    398 			m_adj(m1->m_next, adjlen);
    399 
    400 			/* The second mbuf is guaranteed not to have a pkthdr... */
    401 			m->m_pkthdr.len -= adjlen;
    402 		}
    403 
    404 		/* Now, let's unlink the mbuf chain for a second...*/
    405 		mo = m1->m_next;
    406 		m1->m_next = NULL;
    407 
    408 		/* ...and trim the end of the first part of the chain...sick */
    409 		adjlen = m1->m_len - roff;
    410 		m_adj(m1, -adjlen);
    411 		if (m1 != m)
    412 			m->m_pkthdr.len -= adjlen;
    413 
    414 		/* Finally, let's relink */
    415 		m1->m_next = mo;
    416 	} else {
    417 		/*
    418 		 * The header lies in the "middle" of the mbuf; copy
    419 		 * the remainder of the mbuf down over the header.
    420 		 */
    421 		IPSEC_STATINC(IPSEC_STAT_INPUT_MIDDLE);
    422 		memmove(mtod(m1, u_char *) + roff,
    423 		      mtod(m1, u_char *) + roff + hlen,
    424 		      m1->m_len - (roff + hlen));
    425 		m1->m_len -= hlen;
    426 		m->m_pkthdr.len -= hlen;
    427 	}
    428 
    429 	return 0;
    430 }
    431