1 /* $NetBSD: ipsec_mbuf.c,v 1.5 2004/05/07 00:55:15 jonathan 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 1.5 2004/05/07 00:55:15 jonathan 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 52 #include <netipsec/ipsec_osdep.h> 53 #include <net/net_osdep.h> 54 55 extern struct mbuf *m_getptr(struct mbuf *, int, int *); 56 57 /* 58 * Create a writable copy of the mbuf chain. While doing this 59 * we compact the chain with a goal of producing a chain with 60 * at most two mbufs. The second mbuf in this chain is likely 61 * to be a cluster. The primary purpose of this work is to create 62 * a writable packet for encryption, compression, etc. The 63 * secondary goal is to linearize the data so the data can be 64 * passed to crypto hardware in the most efficient manner possible. 65 */ 66 struct mbuf * 67 m_clone(struct mbuf *m0) 68 { 69 struct mbuf *m, *mprev; 70 struct mbuf *n, *mfirst, *mlast; 71 int len, off; 72 73 IPSEC_ASSERT(m0 != NULL, ("m_clone: null mbuf")); 74 75 mprev = NULL; 76 for (m = m0; m != NULL; m = mprev->m_next) { 77 /* 78 * Regular mbufs are ignored unless there's a cluster 79 * in front of it that we can use to coalesce. We do 80 * the latter mainly so later clusters can be coalesced 81 * also w/o having to handle them specially (i.e. convert 82 * mbuf+cluster -> cluster). This optimization is heavily 83 * influenced by the assumption that we're running over 84 * Ethernet where MCLBYTES is large enough that the max 85 * packet size will permit lots of coalescing into a 86 * single cluster. This in turn permits efficient 87 * crypto operations, especially when using hardware. 88 */ 89 if ((m->m_flags & M_EXT) == 0) { 90 if (mprev && (mprev->m_flags & M_EXT) && 91 m->m_len <= M_TRAILINGSPACE(mprev)) { 92 /* XXX: this ignores mbuf types */ 93 memcpy(mtod(mprev, caddr_t) + mprev->m_len, 94 mtod(m, caddr_t), m->m_len); 95 mprev->m_len += m->m_len; 96 mprev->m_next = m->m_next; /* unlink from chain */ 97 m_free(m); /* reclaim mbuf */ 98 newipsecstat.ips_mbcoalesced++; 99 } else { 100 mprev = m; 101 } 102 continue; 103 } 104 /* 105 * Writable mbufs are left alone (for now). Note 106 * that for 4.x systems it's not possible to identify 107 * whether or not mbufs with external buffers are 108 * writable unless they use clusters. 109 */ 110 if (M_EXT_WRITABLE(m)) { 111 mprev = m; 112 continue; 113 } 114 115 /* 116 * Not writable, replace with a copy or coalesce with 117 * the previous mbuf if possible (since we have to copy 118 * it anyway, we try to reduce the number of mbufs and 119 * clusters so that future work is easier). 120 */ 121 IPSEC_ASSERT(m->m_flags & M_EXT, 122 ("m_clone: m_flags 0x%x", m->m_flags)); 123 /* NB: we only coalesce into a cluster or larger */ 124 if (mprev != NULL && (mprev->m_flags & M_EXT) && 125 m->m_len <= M_TRAILINGSPACE(mprev)) { 126 /* XXX: this ignores mbuf types */ 127 memcpy(mtod(mprev, caddr_t) + mprev->m_len, 128 mtod(m, caddr_t), m->m_len); 129 mprev->m_len += m->m_len; 130 mprev->m_next = m->m_next; /* unlink from chain */ 131 m_free(m); /* reclaim mbuf */ 132 newipsecstat.ips_clcoalesced++; 133 continue; 134 } 135 136 /* 137 * Allocate new space to hold the copy... 138 */ 139 /* XXX why can M_PKTHDR be set past the first mbuf? */ 140 if (mprev == NULL && (m->m_flags & M_PKTHDR)) { 141 /* 142 * NB: if a packet header is present we must 143 * allocate the mbuf separately from any cluster 144 * because M_MOVE_PKTHDR will smash the data 145 * pointer and drop the M_EXT marker. 146 */ 147 MGETHDR(n, M_DONTWAIT, m->m_type); 148 if (n == NULL) { 149 m_freem(m0); 150 return (NULL); 151 } 152 M_MOVE_PKTHDR(n, m); 153 MCLGET(n, M_DONTWAIT); 154 if ((n->m_flags & M_EXT) == 0) { 155 m_free(n); 156 m_freem(m0); 157 return (NULL); 158 } 159 } else { 160 n = m_getcl(M_DONTWAIT, m->m_type, m->m_flags); 161 if (n == NULL) { 162 m_freem(m0); 163 return (NULL); 164 } 165 } 166 /* 167 * ... and copy the data. We deal with jumbo mbufs 168 * (i.e. m_len > MCLBYTES) by splitting them into 169 * clusters. We could just malloc a buffer and make 170 * it external but too many device drivers don't know 171 * how to break up the non-contiguous memory when 172 * doing DMA. 173 */ 174 len = m->m_len; 175 off = 0; 176 mfirst = n; 177 mlast = NULL; 178 for (;;) { 179 int cc = min(len, MCLBYTES); 180 memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + off, cc); 181 n->m_len = cc; 182 if (mlast != NULL) 183 mlast->m_next = n; 184 mlast = n; 185 newipsecstat.ips_clcopied++; 186 187 len -= cc; 188 if (len <= 0) 189 break; 190 off += cc; 191 192 n = m_getcl(M_DONTWAIT, m->m_type, m->m_flags); 193 if (n == NULL) { 194 m_freem(mfirst); 195 m_freem(m0); 196 return (NULL); 197 } 198 } 199 n->m_next = m->m_next; 200 if (mprev == NULL) 201 m0 = mfirst; /* new head of chain */ 202 else 203 mprev->m_next = mfirst; /* replace old mbuf */ 204 m_free(m); /* release old mbuf */ 205 mprev = mfirst; 206 } 207 return (m0); 208 } 209 210 /* 211 * Make space for a new header of length hlen at skip bytes 212 * into the packet. When doing this we allocate new mbufs only 213 * when absolutely necessary. The mbuf where the new header 214 * is to go is returned together with an offset into the mbuf. 215 * If NULL is returned then the mbuf chain may have been modified; 216 * the caller is assumed to always free the chain. 217 */ 218 struct mbuf * 219 m_makespace(struct mbuf *m0, int skip, int hlen, int *off) 220 { 221 struct mbuf *m; 222 unsigned remain; 223 224 IPSEC_ASSERT(m0 != NULL, ("m_dmakespace: null mbuf")); 225 IPSEC_ASSERT(hlen < MHLEN, ("m_makespace: hlen too big: %u", hlen)); 226 227 for (m = m0; m && skip > m->m_len; m = m->m_next) 228 skip -= m->m_len; 229 if (m == NULL) 230 return (NULL); 231 /* 232 * At this point skip is the offset into the mbuf m 233 * where the new header should be placed. Figure out 234 * if there's space to insert the new header. If so, 235 * and copying the remainder makese sense then do so. 236 * Otherwise insert a new mbuf in the chain, splitting 237 * the contents of m as needed. 238 */ 239 remain = m->m_len - skip; /* data to move */ 240 if (hlen > M_TRAILINGSPACE(m)) { 241 struct mbuf *n; 242 243 /* XXX code doesn't handle clusters XXX */ 244 IPSEC_ASSERT(remain < MLEN, 245 ("m_makespace: remainder too big: %u", remain)); 246 /* 247 * Not enough space in m, split the contents 248 * of m, inserting new mbufs as required. 249 * 250 * NB: this ignores mbuf types. 251 */ 252 MGET(n, M_DONTWAIT, MT_DATA); 253 if (n == NULL) 254 return (NULL); 255 n->m_next = m->m_next; /* splice new mbuf */ 256 m->m_next = n; 257 newipsecstat.ips_mbinserted++; 258 if (hlen <= M_TRAILINGSPACE(m) + remain) { 259 /* 260 * New header fits in the old mbuf if we copy 261 * the remainder; just do the copy to the new 262 * mbuf and we're good to go. 263 */ 264 memcpy(mtod(n, caddr_t), 265 mtod(m, caddr_t) + skip, remain); 266 n->m_len = remain; 267 m->m_len = skip + hlen; 268 *off = skip; 269 } else { 270 /* 271 * No space in the old mbuf for the new header. 272 * Make space in the new mbuf and check the 273 * remainder'd data fits too. If not then we 274 * must allocate an additional mbuf (yech). 275 */ 276 n->m_len = 0; 277 if (remain + hlen > M_TRAILINGSPACE(n)) { 278 struct mbuf *n2; 279 280 MGET(n2, M_DONTWAIT, MT_DATA); 281 /* NB: new mbuf is on chain, let caller free */ 282 if (n2 == NULL) 283 return (NULL); 284 n2->m_len = 0; 285 memcpy(mtod(n2, caddr_t), 286 mtod(m, caddr_t) + skip, remain); 287 n2->m_len = remain; 288 /* splice in second mbuf */ 289 n2->m_next = n->m_next; 290 n->m_next = n2; 291 newipsecstat.ips_mbinserted++; 292 } else { 293 memcpy(mtod(n, caddr_t) + hlen, 294 mtod(m, caddr_t) + skip, remain); 295 n->m_len += remain; 296 } 297 m->m_len -= remain; 298 n->m_len += hlen; 299 m = n; /* header is at front ... */ 300 *off = 0; /* ... of new mbuf */ 301 } 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, caddr_t) + skip, 309 mtod(m, caddr_t) + 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 */ 322 caddr_t 323 m_pad(struct mbuf *m, int n) 324 { 325 register struct mbuf *m0, *m1; 326 register int len, pad; 327 caddr_t 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) { 340 IPSEC_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 */ 394 int 395 m_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 newipsecstat.ips_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 newipsecstat.ips_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 newipsecstat.ips_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 */ 459 void 460 m_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 caddr_t addr; 465 466 if (m == NULL) 467 return; 468 printf("%s (off %u len %u): ", where, off, len); 469 addr = mtod(m, caddr_t) + 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, caddr_t) : NULL; 487 } while (m && len > 0); 488 for (m = m0; m; m = m->m_next) 489 printf(" [%p:%u]", mtod(m, caddr_t), m->m_len); 490 printf("\n"); 491 } 492