1 /* $NetBSD: ipsec_mbuf.c,v 1.27 2018/04/28 14:39:34 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 1.27 2018/04/28 14:39:34 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. We do 69 * the latter mainly so later clusters can be coalesced 70 * also w/o having to handle them specially (i.e. convert 71 * mbuf+cluster -> cluster). This optimization is heavily 72 * influenced by the assumption that we're running over 73 * Ethernet where MCLBYTES is large enough that the max 74 * packet size will permit lots of coalescing into a 75 * single cluster. This in turn permits efficient 76 * crypto operations, especially when using hardware. 77 */ 78 if ((m->m_flags & M_EXT) == 0) { 79 if (mprev && (mprev->m_flags & M_EXT) && 80 m->m_len <= M_TRAILINGSPACE(mprev)) { 81 /* XXX: this ignores mbuf types */ 82 memcpy(mtod(mprev, char *) + mprev->m_len, 83 mtod(m, char *), m->m_len); 84 mprev->m_len += m->m_len; 85 mprev->m_next = m->m_next; /* unlink from chain */ 86 m_free(m); /* reclaim mbuf */ 87 IPSEC_STATINC(IPSEC_STAT_MBCOALESCED); 88 } else { 89 mprev = m; 90 } 91 continue; 92 } 93 94 /* 95 * Writable mbufs are left alone (for now). 96 */ 97 if (!M_READONLY(m)) { 98 mprev = m; 99 continue; 100 } 101 102 /* 103 * Not writable, replace with a copy or coalesce with 104 * the previous mbuf if possible (since we have to copy 105 * it anyway, we try to reduce the number of mbufs and 106 * clusters so that future work is easier). 107 */ 108 KASSERTMSG(m->m_flags & M_EXT, "m_flags 0x%x", m->m_flags); 109 /* NB: we only coalesce into a cluster or larger */ 110 if (mprev != NULL && (mprev->m_flags & M_EXT) && 111 m->m_len <= M_TRAILINGSPACE(mprev)) { 112 /* XXX: this ignores mbuf types */ 113 memcpy(mtod(mprev, char *) + mprev->m_len, 114 mtod(m, char *), m->m_len); 115 mprev->m_len += m->m_len; 116 mprev->m_next = m->m_next; /* unlink from chain */ 117 m_free(m); /* reclaim mbuf */ 118 IPSEC_STATINC(IPSEC_STAT_CLCOALESCED); 119 continue; 120 } 121 122 /* 123 * Allocate new space to hold the copy... 124 */ 125 if (mprev == NULL && (m->m_flags & M_PKTHDR)) { 126 /* 127 * NB: if a packet header is present we must 128 * allocate the mbuf separately from any cluster 129 * because M_MOVE_PKTHDR will smash the data 130 * pointer and drop the M_EXT marker. 131 */ 132 MGETHDR(n, M_DONTWAIT, m->m_type); 133 if (n == NULL) { 134 m_freem(m0); 135 return NULL; 136 } 137 M_MOVE_PKTHDR(n, m); 138 MCLGET(n, M_DONTWAIT); 139 if ((n->m_flags & M_EXT) == 0) { 140 m_free(n); 141 m_freem(m0); 142 return NULL; 143 } 144 } else { 145 n = m_getcl(M_DONTWAIT, m->m_type, m->m_flags); 146 if (n == NULL) { 147 m_freem(m0); 148 return NULL; 149 } 150 } 151 152 /* 153 * ... and copy the data. We deal with jumbo mbufs 154 * (i.e. m_len > MCLBYTES) by splitting them into 155 * clusters. We could just malloc a buffer and make 156 * it external but too many device drivers don't know 157 * how to break up the non-contiguous memory when 158 * doing DMA. 159 */ 160 len = m->m_len; 161 off = 0; 162 mfirst = n; 163 mlast = NULL; 164 for (;;) { 165 int cc = min(len, MCLBYTES); 166 memcpy(mtod(n, char *), mtod(m, char *) + off, cc); 167 n->m_len = cc; 168 if (mlast != NULL) 169 mlast->m_next = n; 170 mlast = n; 171 IPSEC_STATINC(IPSEC_STAT_CLCOPIED); 172 173 len -= cc; 174 if (len <= 0) 175 break; 176 off += cc; 177 178 n = m_getcl(M_DONTWAIT, m->m_type, m->m_flags); 179 if (n == NULL) { 180 m_freem(mfirst); 181 m_freem(m0); 182 return NULL; 183 } 184 } 185 n->m_next = m->m_next; 186 if (mprev == NULL) 187 m0 = mfirst; /* new head of chain */ 188 else 189 mprev->m_next = mfirst; /* replace old mbuf */ 190 m_free(m); /* release old mbuf */ 191 mprev = mfirst; 192 } 193 194 return m0; 195 } 196 197 /* 198 * Make space for a new header of length hlen at skip bytes 199 * into the packet. When doing this we allocate new mbufs only 200 * when absolutely necessary. The mbuf where the new header 201 * is to go is returned together with an offset into the mbuf. 202 * If NULL is returned then the mbuf chain may have been modified; 203 * the caller is assumed to always free the chain. 204 */ 205 struct mbuf * 206 m_makespace(struct mbuf *m0, int skip, int hlen, int *off) 207 { 208 struct mbuf *m; 209 unsigned remain; 210 211 KASSERT(m0 != NULL); 212 KASSERT(m0->m_flags & M_PKTHDR); 213 KASSERTMSG(hlen < MHLEN, "hlen too big: %u", hlen); 214 215 for (m = m0; m && skip > m->m_len; m = m->m_next) 216 skip -= m->m_len; 217 if (m == NULL) 218 return NULL; 219 220 /* 221 * At this point skip is the offset into the mbuf m 222 * where the new header should be placed. Figure out 223 * if there's space to insert the new header. If so, 224 * and copying the remainder makes sense then do so. 225 * Otherwise insert a new mbuf in the chain, splitting 226 * the contents of m as needed. 227 */ 228 remain = m->m_len - skip; /* data to move */ 229 if (hlen > M_TRAILINGSPACE(m)) { 230 struct mbuf *n0, *n, **np; 231 int todo, len, done, alloc; 232 233 n0 = NULL; 234 np = &n0; 235 alloc = 0; 236 done = 0; 237 todo = remain; 238 while (todo > 0) { 239 if (todo > MHLEN) { 240 n = m_getcl(M_DONTWAIT, m->m_type, 0); 241 len = MCLBYTES; 242 } else { 243 n = m_get(M_DONTWAIT, m->m_type); 244 len = MHLEN; 245 } 246 if (n == NULL) { 247 m_freem(n0); 248 return NULL; 249 } 250 *np = n; 251 np = &n->m_next; 252 alloc++; 253 len = min(todo, len); 254 memcpy(n->m_data, mtod(m, char *) + skip + done, len); 255 n->m_len = len; 256 done += len; 257 todo -= len; 258 } 259 260 if (hlen <= M_TRAILINGSPACE(m) + remain) { 261 m->m_len = skip + hlen; 262 *off = skip; 263 if (n0 != NULL) { 264 *np = m->m_next; 265 m->m_next = n0; 266 } 267 } else { 268 n = m_get(M_DONTWAIT, m->m_type); 269 if (n == NULL) { 270 m_freem(n0); 271 return NULL; 272 } 273 alloc++; 274 275 if ((n->m_next = n0) == NULL) 276 np = &n->m_next; 277 n0 = n; 278 279 *np = m->m_next; 280 m->m_next = n0; 281 282 n->m_len = hlen; 283 m->m_len = skip; 284 285 m = n; /* header is at front ... */ 286 *off = 0; /* ... of new mbuf */ 287 } 288 289 IPSEC_STATADD(IPSEC_STAT_MBINSERTED, alloc); 290 } else { 291 /* 292 * Copy the remainder to the back of the mbuf 293 * so there's space to write the new header. 294 */ 295 /* XXX can this be memcpy? does it handle overlap? */ 296 memmove(mtod(m, char *) + skip + hlen, 297 mtod(m, char *) + skip, remain); 298 m->m_len += hlen; 299 *off = skip; 300 } 301 302 m0->m_pkthdr.len += hlen; /* adjust packet length */ 303 return m; 304 } 305 306 /* 307 * m_pad(m, n) pads <m> with <n> bytes at the end. The packet header 308 * length is updated, and a pointer to the first byte of the padding 309 * (which is guaranteed to be all in one mbuf) is returned. 310 */ 311 void * 312 m_pad(struct mbuf *m, int n) 313 { 314 register struct mbuf *m0, *m1; 315 register int len, pad; 316 void *retval; 317 318 if (__predict_false(n > MLEN)) { 319 panic("%s: %d > MLEN", __func__, n); 320 } 321 KASSERT(m->m_flags & M_PKTHDR); 322 323 len = m->m_pkthdr.len; 324 pad = n; 325 m0 = m; 326 327 while (m0->m_len < len) { 328 KASSERTMSG(m0->m_next != NULL, 329 "m0 null, len %u m_len %u", len, m0->m_len); 330 len -= m0->m_len; 331 m0 = m0->m_next; 332 } 333 334 if (m0->m_len != len) { 335 IPSECLOG(LOG_DEBUG, 336 "length mismatch (should be %d instead of %d)\n", 337 m->m_pkthdr.len, m->m_pkthdr.len + m0->m_len - len); 338 m_freem(m); 339 return NULL; 340 } 341 342 /* Check for zero-length trailing mbufs, and find the last one. */ 343 for (m1 = m0; m1->m_next; m1 = m1->m_next) { 344 if (m1->m_next->m_len != 0) { 345 IPSECLOG(LOG_DEBUG, 346 "length mismatch (should be %d instead of %d)\n", 347 m->m_pkthdr.len, 348 m->m_pkthdr.len + m1->m_next->m_len); 349 m_freem(m); 350 return NULL; 351 } 352 353 m0 = m1->m_next; 354 } 355 356 if (pad > M_TRAILINGSPACE(m0)) { 357 /* Add an mbuf to the chain. */ 358 MGET(m1, M_DONTWAIT, MT_DATA); 359 if (m1 == NULL) { 360 m_freem(m); 361 IPSECLOG(LOG_DEBUG, "unable to get extra mbuf\n"); 362 return NULL; 363 } 364 365 m0->m_next = m1; 366 m0 = m1; 367 m0->m_len = 0; 368 } 369 370 retval = m0->m_data + m0->m_len; 371 m0->m_len += pad; 372 m->m_pkthdr.len += pad; 373 374 return retval; 375 } 376 377 /* 378 * Remove hlen data at offset skip in the packet. This is used by 379 * the protocols strip protocol headers and associated data (e.g. IV, 380 * authenticator) on input. 381 */ 382 int 383 m_striphdr(struct mbuf *m, int skip, int hlen) 384 { 385 struct mbuf *m1; 386 int roff; 387 388 KASSERT(m->m_flags & M_PKTHDR); 389 390 /* Find beginning of header */ 391 m1 = m_getptr(m, skip, &roff); 392 if (m1 == NULL) 393 return EINVAL; 394 395 /* Remove the header and associated data from the mbuf. */ 396 if (roff == 0) { 397 /* The header was at the beginning of the mbuf */ 398 IPSEC_STATINC(IPSEC_STAT_INPUT_FRONT); 399 m_adj(m1, hlen); 400 if (m1 != m) 401 m->m_pkthdr.len -= hlen; 402 } else if (roff + hlen >= m1->m_len) { 403 struct mbuf *mo; 404 int adjlen; 405 406 /* 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. 410 */ 411 IPSEC_STATINC(IPSEC_STAT_INPUT_END); 412 if (roff + hlen > m1->m_len) { 413 adjlen = roff + hlen - m1->m_len; 414 415 /* Adjust the next mbuf by the remainder */ 416 m_adj(m1->m_next, adjlen); 417 418 /* The second mbuf is guaranteed not to have a pkthdr... */ 419 m->m_pkthdr.len -= adjlen; 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 adjlen = m1->m_len - roff; 428 m_adj(m1, -adjlen); 429 if (m1 != m) 430 m->m_pkthdr.len -= adjlen; 431 432 /* Finally, let's relink */ 433 m1->m_next = mo; 434 } else { 435 /* 436 * The header lies in the "middle" of the mbuf; copy 437 * the remainder of the mbuf down over the header. 438 */ 439 IPSEC_STATINC(IPSEC_STAT_INPUT_MIDDLE); 440 memmove(mtod(m1, u_char *) + roff, 441 mtod(m1, u_char *) + roff + hlen, 442 m1->m_len - (roff + hlen)); 443 m1->m_len -= hlen; 444 m->m_pkthdr.len -= hlen; 445 } 446 447 return 0; 448 } 449