ipsec_mbuf.c revision 1.22
1/* $NetBSD: ipsec_mbuf.c,v 1.22 2018/03/10 17:52:50 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: /repoman/r/ncvs/src/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 1.22 2018/03/10 17:52:50 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#include <sys/socket.h> 42 43#include <net/route.h> 44#include <netinet/in.h> 45 46#include <netipsec/ipsec.h> 47#include <netipsec/ipsec_var.h> 48#include <netipsec/ipsec_private.h> 49 50/* 51 * Create a writable copy of the mbuf chain. While doing this 52 * we compact the chain with a goal of producing a chain with 53 * at most two mbufs. The second mbuf in this chain is likely 54 * to be a cluster. The primary purpose of this work is to create 55 * a writable packet for encryption, compression, etc. The 56 * secondary goal is to linearize the data so the data can be 57 * passed to crypto hardware in the most efficient manner possible. 58 */ 59struct mbuf * 60m_clone(struct mbuf *m0) 61{ 62 struct mbuf *m, *mprev; 63 struct mbuf *n, *mfirst, *mlast; 64 int len, off; 65 66 KASSERT(m0 != NULL); 67 68 mprev = NULL; 69 for (m = m0; m != NULL; m = mprev->m_next) { 70 /* 71 * Regular mbufs are ignored unless there's a cluster 72 * in front of it that we can use to coalesce. We do 73 * the latter mainly so later clusters can be coalesced 74 * also w/o having to handle them specially (i.e. convert 75 * mbuf+cluster -> cluster). This optimization is heavily 76 * influenced by the assumption that we're running over 77 * Ethernet where MCLBYTES is large enough that the max 78 * packet size will permit lots of coalescing into a 79 * single cluster. This in turn permits efficient 80 * crypto operations, especially when using hardware. 81 */ 82 if ((m->m_flags & M_EXT) == 0) { 83 if (mprev && (mprev->m_flags & M_EXT) && 84 m->m_len <= M_TRAILINGSPACE(mprev)) { 85 /* XXX: this ignores mbuf types */ 86 memcpy(mtod(mprev, char *) + mprev->m_len, 87 mtod(m, char *), m->m_len); 88 mprev->m_len += m->m_len; 89 mprev->m_next = m->m_next; /* unlink from chain */ 90 m_free(m); /* reclaim mbuf */ 91 IPSEC_STATINC(IPSEC_STAT_MBCOALESCED); 92 } else { 93 mprev = m; 94 } 95 continue; 96 } 97 /* 98 * Writable mbufs are left alone (for now). Note 99 * that for 4.x systems it's not possible to identify 100 * whether or not mbufs with external buffers are 101 * writable unless they use clusters. 102 */ 103 if (M_EXT_WRITABLE(m)) { 104 mprev = m; 105 continue; 106 } 107 108 /* 109 * Not writable, replace with a copy or coalesce with 110 * the previous mbuf if possible (since we have to copy 111 * it anyway, we try to reduce the number of mbufs and 112 * clusters so that future work is easier). 113 */ 114 KASSERTMSG(m->m_flags & M_EXT, "m_flags 0x%x", m->m_flags); 115 /* NB: we only coalesce into a cluster or larger */ 116 if (mprev != NULL && (mprev->m_flags & M_EXT) && 117 m->m_len <= M_TRAILINGSPACE(mprev)) { 118 /* XXX: this ignores mbuf types */ 119 memcpy(mtod(mprev, char *) + mprev->m_len, 120 mtod(m, char *), m->m_len); 121 mprev->m_len += m->m_len; 122 mprev->m_next = m->m_next; /* unlink from chain */ 123 m_free(m); /* reclaim mbuf */ 124 IPSEC_STATINC(IPSEC_STAT_CLCOALESCED); 125 continue; 126 } 127 128 /* 129 * Allocate new space to hold the copy... 130 */ 131 /* XXX why can M_PKTHDR be set past the first mbuf? */ 132 if (mprev == NULL && (m->m_flags & M_PKTHDR)) { 133 /* 134 * NB: if a packet header is present we must 135 * allocate the mbuf separately from any cluster 136 * because M_MOVE_PKTHDR will smash the data 137 * pointer and drop the M_EXT marker. 138 */ 139 MGETHDR(n, M_DONTWAIT, m->m_type); 140 if (n == NULL) { 141 m_freem(m0); 142 return (NULL); 143 } 144 M_MOVE_PKTHDR(n, m); 145 MCLGET(n, M_DONTWAIT); 146 if ((n->m_flags & M_EXT) == 0) { 147 m_free(n); 148 m_freem(m0); 149 return (NULL); 150 } 151 } else { 152 n = m_getcl(M_DONTWAIT, m->m_type, m->m_flags); 153 if (n == NULL) { 154 m_freem(m0); 155 return (NULL); 156 } 157 } 158 /* 159 * ... and copy the data. We deal with jumbo mbufs 160 * (i.e. m_len > MCLBYTES) by splitting them into 161 * clusters. We could just malloc a buffer and make 162 * it external but too many device drivers don't know 163 * how to break up the non-contiguous memory when 164 * doing DMA. 165 */ 166 len = m->m_len; 167 off = 0; 168 mfirst = n; 169 mlast = NULL; 170 for (;;) { 171 int cc = min(len, MCLBYTES); 172 memcpy(mtod(n, char *), mtod(m, char *) + off, cc); 173 n->m_len = cc; 174 if (mlast != NULL) 175 mlast->m_next = n; 176 mlast = n; 177 IPSEC_STATINC(IPSEC_STAT_CLCOPIED); 178 179 len -= cc; 180 if (len <= 0) 181 break; 182 off += cc; 183 184 n = m_getcl(M_DONTWAIT, m->m_type, m->m_flags); 185 if (n == NULL) { 186 m_freem(mfirst); 187 m_freem(m0); 188 return (NULL); 189 } 190 } 191 n->m_next = m->m_next; 192 if (mprev == NULL) 193 m0 = mfirst; /* new head of chain */ 194 else 195 mprev->m_next = mfirst; /* replace old mbuf */ 196 m_free(m); /* release old mbuf */ 197 mprev = mfirst; 198 } 199 return (m0); 200} 201 202/* 203 * Make space for a new header of length hlen at skip bytes 204 * into the packet. When doing this we allocate new mbufs only 205 * when absolutely necessary. The mbuf where the new header 206 * is to go is returned together with an offset into the mbuf. 207 * If NULL is returned then the mbuf chain may have been modified; 208 * the caller is assumed to always free the chain. 209 */ 210struct mbuf * 211m_makespace(struct mbuf *m0, int skip, int hlen, int *off) 212{ 213 struct mbuf *m; 214 unsigned remain; 215 216 KASSERT(m0 != NULL); 217 KASSERT(m0->m_flags & M_PKTHDR); 218 KASSERTMSG(hlen < MHLEN, "hlen too big: %u", hlen); 219 220 for (m = m0; m && skip > m->m_len; m = m->m_next) 221 skip -= m->m_len; 222 if (m == NULL) 223 return (NULL); 224 /* 225 * At this point skip is the offset into the mbuf m 226 * where the new header should be placed. Figure out 227 * if there's space to insert the new header. If so, 228 * and copying the remainder makes sense then do so. 229 * Otherwise insert a new mbuf in the chain, splitting 230 * the contents of m as needed. 231 */ 232 remain = m->m_len - skip; /* data to move */ 233 if (hlen > M_TRAILINGSPACE(m)) { 234 struct mbuf *n0, *n, **np; 235 int todo, len, done, alloc; 236 237 n0 = NULL; 238 np = &n0; 239 alloc = 0; 240 done = 0; 241 todo = remain; 242 while (todo > 0) { 243 if (todo > MHLEN) { 244 n = m_getcl(M_DONTWAIT, m->m_type, 0); 245 len = MCLBYTES; 246 } else { 247 n = m_get(M_DONTWAIT, m->m_type); 248 len = MHLEN; 249 } 250 if (n == NULL) { 251 m_freem(n0); 252 return NULL; 253 } 254 *np = n; 255 np = &n->m_next; 256 alloc++; 257 len = min(todo, len); 258 memcpy(n->m_data, mtod(m, char *) + skip + done, len); 259 n->m_len = len; 260 done += len; 261 todo -= len; 262 } 263 264 if (hlen <= M_TRAILINGSPACE(m) + remain) { 265 m->m_len = skip + hlen; 266 *off = skip; 267 if (n0 != NULL) { 268 *np = m->m_next; 269 m->m_next = n0; 270 } 271 } else { 272 n = m_get(M_DONTWAIT, m->m_type); 273 if (n == NULL) { 274 m_freem(n0); 275 return NULL; 276 } 277 alloc++; 278 279 if ((n->m_next = n0) == NULL) 280 np = &n->m_next; 281 n0 = n; 282 283 *np = m->m_next; 284 m->m_next = n0; 285 286 n->m_len = hlen; 287 m->m_len = skip; 288 289 m = n; /* header is at front ... */ 290 *off = 0; /* ... of new mbuf */ 291 } 292 293 IPSEC_STATADD(IPSEC_STAT_MBINSERTED, alloc); 294 } else { 295 /* 296 * Copy the remainder to the back of the mbuf 297 * so there's space to write the new header. 298 */ 299 /* XXX can this be memcpy? does it handle overlap? */ 300 memmove(mtod(m, char *) + skip + hlen, 301 mtod(m, char *) + skip, remain); 302 m->m_len += hlen; 303 *off = skip; 304 } 305 m0->m_pkthdr.len += hlen; /* adjust packet length */ 306 return m; 307} 308 309/* 310 * m_pad(m, n) pads <m> with <n> bytes at the end. The packet header 311 * length is updated, and a pointer to the first byte of the padding 312 * (which is guaranteed to be all in one mbuf) is returned. 313 */ 314void * 315m_pad(struct mbuf *m, int n) 316{ 317 register struct mbuf *m0, *m1; 318 register int len, pad; 319 void *retval; 320 321 if (__predict_false(n > MLEN)) { 322 panic("%s: %d > MLEN", __func__, n); 323 } 324 KASSERT(m->m_flags & M_PKTHDR); 325 326 len = m->m_pkthdr.len; 327 pad = n; 328 m0 = m; 329 330 while (m0->m_len < len) { 331 KASSERTMSG(m0->m_next != NULL, 332 "m0 null, len %u m_len %u", len, m0->m_len); 333 len -= m0->m_len; 334 m0 = m0->m_next; 335 } 336 337 if (m0->m_len != len) { 338 IPSECLOG(LOG_DEBUG, 339 "length mismatch (should be %d instead of %d)\n", 340 m->m_pkthdr.len, m->m_pkthdr.len + m0->m_len - len); 341 m_freem(m); 342 return NULL; 343 } 344 345 /* Check for zero-length trailing mbufs, and find the last one. */ 346 for (m1 = m0; m1->m_next; m1 = m1->m_next) { 347 if (m1->m_next->m_len != 0) { 348 IPSECLOG(LOG_DEBUG, 349 "length mismatch (should be %d instead of %d)\n", 350 m->m_pkthdr.len, 351 m->m_pkthdr.len + m1->m_next->m_len); 352 m_freem(m); 353 return NULL; 354 } 355 356 m0 = m1->m_next; 357 } 358 359 if (pad > M_TRAILINGSPACE(m0)) { 360 /* Add an mbuf to the chain. */ 361 MGET(m1, M_DONTWAIT, MT_DATA); 362 if (m1 == NULL) { 363 m_freem(m); 364 IPSECLOG(LOG_DEBUG, "unable to get extra mbuf\n"); 365 return NULL; 366 } 367 368 m0->m_next = m1; 369 m0 = m1; 370 m0->m_len = 0; 371 } 372 373 retval = m0->m_data + m0->m_len; 374 m0->m_len += pad; 375 m->m_pkthdr.len += pad; 376 377 return retval; 378} 379 380/* 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. 384 */ 385int 386m_striphdr(struct mbuf *m, int skip, int hlen) 387{ 388 struct mbuf *m1; 389 int roff; 390 391 KASSERT(m->m_flags & M_PKTHDR); 392 393 /* Find beginning of header */ 394 m1 = m_getptr(m, skip, &roff); 395 if (m1 == NULL) 396 return (EINVAL); 397 398 /* Remove the header and associated data from the mbuf. */ 399 if (roff == 0) { 400 /* The header was at the beginning of the mbuf */ 401 IPSEC_STATINC(IPSEC_STAT_INPUT_FRONT); 402 m_adj(m1, hlen); 403 if ((m1->m_flags & M_PKTHDR) == 0) 404 m->m_pkthdr.len -= hlen; 405 } else if (roff + hlen >= m1->m_len) { 406 struct mbuf *mo; 407 408 /* 409 * Part or all of the header is at the end of this mbuf, 410 * so first let's remove the remainder of the header from 411 * the beginning of the remainder of the mbuf chain, if any. 412 */ 413 IPSEC_STATINC(IPSEC_STAT_INPUT_END); 414 if (roff + hlen > m1->m_len) { 415 /* Adjust the next mbuf by the remainder */ 416 m_adj(m1->m_next, roff + hlen - m1->m_len); 417 418 /* The second mbuf is guaranteed not to have a pkthdr... */ 419 m->m_pkthdr.len -= (roff + hlen - m1->m_len); 420 } 421 422 /* Now, let's unlink the mbuf chain for a second...*/ 423 mo = m1->m_next; 424 m1->m_next = NULL; 425 426 /* ...and trim the end of the first part of the chain...sick */ 427 m_adj(m1, -(m1->m_len - roff)); 428 if ((m1->m_flags & M_PKTHDR) == 0) 429 m->m_pkthdr.len -= (m1->m_len - roff); 430 431 /* Finally, let's relink */ 432 m1->m_next = mo; 433 } else { 434 /* 435 * The header lies in the "middle" of the mbuf; copy 436 * the remainder of the mbuf down over the header. 437 */ 438 IPSEC_STATINC(IPSEC_STAT_INPUT_MIDDLE); 439 memmove(mtod(m1, u_char *) + roff, 440 mtod(m1, u_char *) + roff + hlen, 441 m1->m_len - (roff + hlen)); 442 m1->m_len -= hlen; 443 m->m_pkthdr.len -= hlen; 444 } 445 return (0); 446} 447