1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 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. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)uipc_mbuf.c 8.2 (Berkeley) 1/4/94 30 */ 31 32 #include <sys/cdefs.h> 33 __FBSDID("$FreeBSD$"); 34 35 #include "opt_mac.h" 36 #include "opt_param.h" 37 #include "opt_mbuf_stress_test.h" 38 39 #include <sys/param.h> 40 #include <sys/systm.h> 41 #include <sys/kernel.h> 42 #include <sys/limits.h> 43 #include <sys/lock.h> 44 #include <sys/malloc.h> 45 #include <sys/mbuf.h> 46 #include <sys/sysctl.h> 47 #include <sys/domain.h> 48 #include <sys/protosw.h> 49 #include <sys/uio.h> 50 51 #include <security/mac/mac_framework.h> 52 53 int max_linkhdr; 54 int max_protohdr; 55 int max_hdr; 56 int max_datalen; 57 #ifdef MBUF_STRESS_TEST 58 int m_defragpackets; 59 int m_defragbytes; 60 int m_defraguseless; 61 int m_defragfailure; 62 int m_defragrandomfailures; 63 #endif 64 65 /* 66 * sysctl(8) exported objects 67 */ 68 SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RD, 69 &max_linkhdr, 0, "Size of largest link layer header"); 70 SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RD, 71 &max_protohdr, 0, "Size of largest protocol layer header"); 72 SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RD, 73 &max_hdr, 0, "Size of largest link plus protocol header"); 74 SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RD, 75 &max_datalen, 0, "Minimum space left in mbuf after max_hdr"); 76 #ifdef MBUF_STRESS_TEST 77 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD, 78 &m_defragpackets, 0, ""); 79 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD, 80 &m_defragbytes, 0, ""); 81 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD, 82 &m_defraguseless, 0, ""); 83 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD, 84 &m_defragfailure, 0, ""); 85 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW, 86 &m_defragrandomfailures, 0, ""); 87 #endif 88 89 /* 90 * Allocate a given length worth of mbufs and/or clusters (whatever fits 91 * best) and return a pointer to the top of the allocated chain. If an 92 * existing mbuf chain is provided, then we will append the new chain 93 * to the existing one but still return the top of the newly allocated 94 * chain. 95 */ 96 struct mbuf * 97 m_getm2(struct mbuf *m, int len, int how, short type, int flags) 98 { 99 struct mbuf *mb, *nm = NULL, *mtail = NULL; 100 101 KASSERT(len >= 0, ("%s: len is < 0", __func__)); 102 103 /* Validate flags. */ 104 flags &= (M_PKTHDR | M_EOR); 105 106 /* Packet header mbuf must be first in chain. */ 107 if ((flags & M_PKTHDR) && m != NULL) 108 flags &= ~M_PKTHDR; 109 110 /* Loop and append maximum sized mbufs to the chain tail. */ 111 while (len > 0) { 112 if (len > MCLBYTES) 113 mb = m_getjcl(how, type, (flags & M_PKTHDR), 114 MJUMPAGESIZE); 115 else if (len >= MINCLSIZE) 116 mb = m_getcl(how, type, (flags & M_PKTHDR)); 117 else if (flags & M_PKTHDR) 118 mb = m_gethdr(how, type); 119 else 120 mb = m_get(how, type); 121 122 /* Fail the whole operation if one mbuf can't be allocated. */ 123 if (mb == NULL) { 124 if (nm != NULL) 125 m_freem(nm); 126 return (NULL); 127 } 128 129 /* Book keeping. */ 130 len -= (mb->m_flags & M_EXT) ? mb->m_ext.ext_size : 131 ((mb->m_flags & M_PKTHDR) ? MHLEN : MLEN); 132 if (mtail != NULL) 133 mtail->m_next = mb; 134 else 135 nm = mb; 136 mtail = mb; 137 flags &= ~M_PKTHDR; /* Only valid on the first mbuf. */ 138 } 139 if (flags & M_EOR) 140 mtail->m_flags |= M_EOR; /* Only valid on the last mbuf. */ 141 142 /* If mbuf was supplied, append new chain to the end of it. */ 143 if (m != NULL) { 144 for (mtail = m; mtail->m_next != NULL; mtail = mtail->m_next) 145 ; 146 mtail->m_next = nm; 147 mtail->m_flags &= ~M_EOR; 148 } else 149 m = nm; 150 151 return (m); 152 } 153 154 /* 155 * Free an entire chain of mbufs and associated external buffers, if 156 * applicable. 157 */ 158 void 159 m_freem(struct mbuf *mb) 160 { 161 162 while (mb != NULL) 163 mb = m_free(mb); 164 } 165 166 /*- 167 * Configure a provided mbuf to refer to the provided external storage 168 * buffer and setup a reference count for said buffer. If the setting 169 * up of the reference count fails, the M_EXT bit will not be set. If 170 * successfull, the M_EXT bit is set in the mbuf's flags. 171 * 172 * Arguments: 173 * mb The existing mbuf to which to attach the provided buffer. 174 * buf The address of the provided external storage buffer. 175 * size The size of the provided buffer. 176 * freef A pointer to a routine that is responsible for freeing the 177 * provided external storage buffer. 178 * args A pointer to an argument structure (of any type) to be passed 179 * to the provided freef routine (may be NULL). 180 * flags Any other flags to be passed to the provided mbuf. 181 * type The type that the external storage buffer should be 182 * labeled with. 183 * 184 * Returns: 185 * Nothing. 186 */ 187 void 188 m_extadd(struct mbuf *mb, caddr_t buf, u_int size, 189 void (*freef)(void *, void *), void *args, int flags, int type) 190 { 191 KASSERT(type != EXT_CLUSTER, ("%s: EXT_CLUSTER not allowed", __func__)); 192 193 if (type != EXT_EXTREF) 194 mb->m_ext.ref_cnt = (u_int *)uma_zalloc(zone_ext_refcnt, M_NOWAIT); 195 if (mb->m_ext.ref_cnt != NULL) { 196 *(mb->m_ext.ref_cnt) = 1; 197 mb->m_flags |= (M_EXT | flags); 198 mb->m_ext.ext_buf = buf; 199 mb->m_data = mb->m_ext.ext_buf; 200 mb->m_ext.ext_size = size; 201 mb->m_ext.ext_free = freef; 202 mb->m_ext.ext_args = args; 203 mb->m_ext.ext_type = type; 204 } 205 } 206 207 /* 208 * Non-directly-exported function to clean up after mbufs with M_EXT 209 * storage attached to them if the reference count hits 1. 210 */ 211 void 212 mb_free_ext(struct mbuf *m) 213 { 214 int skipmbuf; 215 216 KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__)); 217 KASSERT(m->m_ext.ref_cnt != NULL, ("%s: ref_cnt not set", __func__)); 218 219 220 /* 221 * check if the header is embedded in the cluster 222 */ 223 skipmbuf = (m->m_flags & M_NOFREE); 224 225 /* Free attached storage if this mbuf is the only reference to it. */ 226 if (*(m->m_ext.ref_cnt) == 1 || 227 atomic_fetchadd_int(m->m_ext.ref_cnt, -1) == 1) { 228 switch (m->m_ext.ext_type) { 229 case EXT_PACKET: /* The packet zone is special. */ 230 if (*(m->m_ext.ref_cnt) == 0) 231 *(m->m_ext.ref_cnt) = 1; 232 uma_zfree(zone_pack, m); 233 return; /* Job done. */ 234 case EXT_CLUSTER: 235 uma_zfree(zone_clust, m->m_ext.ext_buf); 236 break; 237 case EXT_JUMBOP: 238 uma_zfree(zone_jumbop, m->m_ext.ext_buf); 239 break; 240 case EXT_JUMBO9: 241 uma_zfree(zone_jumbo9, m->m_ext.ext_buf); 242 break; 243 case EXT_JUMBO16: 244 uma_zfree(zone_jumbo16, m->m_ext.ext_buf); 245 break; 246 case EXT_SFBUF: 247 case EXT_NET_DRV: 248 case EXT_MOD_TYPE: 249 case EXT_DISPOSABLE: 250 *(m->m_ext.ref_cnt) = 0; 251 uma_zfree(zone_ext_refcnt, __DEVOLATILE(u_int *, 252 m->m_ext.ref_cnt)); 253 /* FALLTHROUGH */ 254 case EXT_EXTREF: 255 KASSERT(m->m_ext.ext_free != NULL, 256 ("%s: ext_free not set", __func__)); 257 (*(m->m_ext.ext_free))(m->m_ext.ext_buf, 258 m->m_ext.ext_args); 259 break; 260 default: 261 KASSERT(m->m_ext.ext_type == 0, 262 ("%s: unknown ext_type", __func__)); 263 } 264 } 265 if (skipmbuf) 266 return; 267 268 /* 269 * Free this mbuf back to the mbuf zone with all m_ext 270 * information purged. 271 */ 272 m->m_ext.ext_buf = NULL; 273 m->m_ext.ext_free = NULL; 274 m->m_ext.ext_args = NULL; 275 m->m_ext.ref_cnt = NULL; 276 m->m_ext.ext_size = 0; 277 m->m_ext.ext_type = 0; 278 m->m_flags &= ~M_EXT; 279 uma_zfree(zone_mbuf, m); 280 } 281 282 /* 283 * Attach the the cluster from *m to *n, set up m_ext in *n 284 * and bump the refcount of the cluster. 285 */ 286 static void 287 mb_dupcl(struct mbuf *n, struct mbuf *m) 288 { 289 KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__)); 290 KASSERT(m->m_ext.ref_cnt != NULL, ("%s: ref_cnt not set", __func__)); 291 KASSERT((n->m_flags & M_EXT) == 0, ("%s: M_EXT set", __func__)); 292 293 if (*(m->m_ext.ref_cnt) == 1) 294 *(m->m_ext.ref_cnt) += 1; 295 else 296 atomic_add_int(m->m_ext.ref_cnt, 1); 297 n->m_ext.ext_buf = m->m_ext.ext_buf; 298 n->m_ext.ext_free = m->m_ext.ext_free; 299 n->m_ext.ext_args = m->m_ext.ext_args; 300 n->m_ext.ext_size = m->m_ext.ext_size; 301 n->m_ext.ref_cnt = m->m_ext.ref_cnt; 302 n->m_ext.ext_type = m->m_ext.ext_type; 303 n->m_flags |= M_EXT; 304 } 305 306 /* 307 * Clean up mbuf (chain) from any tags and packet headers. 308 * If "all" is set then the first mbuf in the chain will be 309 * cleaned too. 310 */ 311 void 312 m_demote(struct mbuf *m0, int all) 313 { 314 struct mbuf *m; 315 316 for (m = all ? m0 : m0->m_next; m != NULL; m = m->m_next) { 317 if (m->m_flags & M_PKTHDR) { 318 m_tag_delete_chain(m, NULL); 319 m->m_flags &= ~M_PKTHDR; 320 bzero(&m->m_pkthdr, sizeof(struct pkthdr)); 321 } 322 if (m->m_type == MT_HEADER) 323 m->m_type = MT_DATA; 324 if (m != m0 && m->m_nextpkt != NULL) 325 m->m_nextpkt = NULL; 326 m->m_flags = m->m_flags & (M_EXT|M_EOR|M_RDONLY|M_FREELIST); 327 } 328 } 329 330 /* 331 * Sanity checks on mbuf (chain) for use in KASSERT() and general 332 * debugging. 333 * Returns 0 or panics when bad and 1 on all tests passed. 334 * Sanitize, 0 to run M_SANITY_ACTION, 1 to garble things so they 335 * blow up later. 336 */ 337 int 338 m_sanity(struct mbuf *m0, int sanitize) 339 { 340 struct mbuf *m; 341 caddr_t a, b; 342 int pktlen = 0; 343 344 #ifdef INVARIANTS 345 #define M_SANITY_ACTION(s) panic("mbuf %p: " s, m) 346 #else 347 #define M_SANITY_ACTION(s) printf("mbuf %p: " s, m) 348 #endif 349 350 for (m = m0; m != NULL; m = m->m_next) { 351 /* 352 * Basic pointer checks. If any of these fails then some 353 * unrelated kernel memory before or after us is trashed. 354 * No way to recover from that. 355 */ 356 a = ((m->m_flags & M_EXT) ? m->m_ext.ext_buf : 357 ((m->m_flags & M_PKTHDR) ? (caddr_t)(&m->m_pktdat) : 358 (caddr_t)(&m->m_dat)) ); 359 b = (caddr_t)(a + (m->m_flags & M_EXT ? m->m_ext.ext_size : 360 ((m->m_flags & M_PKTHDR) ? MHLEN : MLEN))); 361 if ((caddr_t)m->m_data < a) 362 M_SANITY_ACTION("m_data outside mbuf data range left"); 363 if ((caddr_t)m->m_data > b) 364 M_SANITY_ACTION("m_data outside mbuf data range right"); 365 if ((caddr_t)m->m_data + m->m_len > b) 366 M_SANITY_ACTION("m_data + m_len exeeds mbuf space"); 367 if ((m->m_flags & M_PKTHDR) && m->m_pkthdr.header) { 368 if ((caddr_t)m->m_pkthdr.header < a || 369 (caddr_t)m->m_pkthdr.header > b) 370 M_SANITY_ACTION("m_pkthdr.header outside mbuf data range"); 371 } 372 373 /* m->m_nextpkt may only be set on first mbuf in chain. */ 374 if (m != m0 && m->m_nextpkt != NULL) { 375 if (sanitize) { 376 m_freem(m->m_nextpkt); 377 m->m_nextpkt = (struct mbuf *)0xDEADC0DE; 378 } else 379 M_SANITY_ACTION("m->m_nextpkt on in-chain mbuf"); 380 } 381 382 /* packet length (not mbuf length!) calculation */ 383 if (m0->m_flags & M_PKTHDR) 384 pktlen += m->m_len; 385 386 /* m_tags may only be attached to first mbuf in chain. */ 387 if (m != m0 && m->m_flags & M_PKTHDR && 388 !SLIST_EMPTY(&m->m_pkthdr.tags)) { 389 if (sanitize) { 390 m_tag_delete_chain(m, NULL); 391 /* put in 0xDEADC0DE perhaps? */ 392 } else 393 M_SANITY_ACTION("m_tags on in-chain mbuf"); 394 } 395 396 /* M_PKTHDR may only be set on first mbuf in chain */ 397 if (m != m0 && m->m_flags & M_PKTHDR) { 398 if (sanitize) { 399 bzero(&m->m_pkthdr, sizeof(m->m_pkthdr)); 400 m->m_flags &= ~M_PKTHDR; 401 /* put in 0xDEADCODE and leave hdr flag in */ 402 } else 403 M_SANITY_ACTION("M_PKTHDR on in-chain mbuf"); 404 } 405 } 406 m = m0; 407 if (pktlen && pktlen != m->m_pkthdr.len) { 408 if (sanitize) 409 m->m_pkthdr.len = 0; 410 else 411 M_SANITY_ACTION("m_pkthdr.len != mbuf chain length"); 412 } 413 return 1; 414 415 #undef M_SANITY_ACTION 416 } 417 418 419 /* 420 * "Move" mbuf pkthdr from "from" to "to". 421 * "from" must have M_PKTHDR set, and "to" must be empty. 422 */ 423 void 424 m_move_pkthdr(struct mbuf *to, struct mbuf *from) 425 { 426 427 #if 0 428 /* see below for why these are not enabled */ 429 M_ASSERTPKTHDR(to); 430 /* Note: with MAC, this may not be a good assertion. */ 431 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), 432 ("m_move_pkthdr: to has tags")); 433 #endif 434 #ifdef MAC 435 /* 436 * XXXMAC: It could be this should also occur for non-MAC? 437 */ 438 if (to->m_flags & M_PKTHDR) 439 m_tag_delete_chain(to, NULL); 440 #endif 441 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT); 442 if ((to->m_flags & M_EXT) == 0) 443 to->m_data = to->m_pktdat; 444 to->m_pkthdr = from->m_pkthdr; /* especially tags */ 445 SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */ 446 from->m_flags &= ~M_PKTHDR; 447 } 448 449 /* 450 * Duplicate "from"'s mbuf pkthdr in "to". 451 * "from" must have M_PKTHDR set, and "to" must be empty. 452 * In particular, this does a deep copy of the packet tags. 453 */ 454 int 455 m_dup_pkthdr(struct mbuf *to, struct mbuf *from, int how) 456 { 457 458 #if 0 459 /* 460 * The mbuf allocator only initializes the pkthdr 461 * when the mbuf is allocated with MGETHDR. Many users 462 * (e.g. m_copy*, m_prepend) use MGET and then 463 * smash the pkthdr as needed causing these 464 * assertions to trip. For now just disable them. 465 */ 466 M_ASSERTPKTHDR(to); 467 /* Note: with MAC, this may not be a good assertion. */ 468 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), ("m_dup_pkthdr: to has tags")); 469 #endif 470 MBUF_CHECKSLEEP(how); 471 #ifdef MAC 472 if (to->m_flags & M_PKTHDR) 473 m_tag_delete_chain(to, NULL); 474 #endif 475 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT); 476 if ((to->m_flags & M_EXT) == 0) 477 to->m_data = to->m_pktdat; 478 to->m_pkthdr = from->m_pkthdr; 479 SLIST_INIT(&to->m_pkthdr.tags); 480 return (m_tag_copy_chain(to, from, MBTOM(how))); 481 } 482 483 /* 484 * Lesser-used path for M_PREPEND: 485 * allocate new mbuf to prepend to chain, 486 * copy junk along. 487 */ 488 struct mbuf * 489 m_prepend(struct mbuf *m, int len, int how) 490 { 491 struct mbuf *mn; 492 493 if (m->m_flags & M_PKTHDR) 494 MGETHDR(mn, how, m->m_type); 495 else 496 MGET(mn, how, m->m_type); 497 if (mn == NULL) { 498 m_freem(m); 499 return (NULL); 500 } 501 if (m->m_flags & M_PKTHDR) 502 M_MOVE_PKTHDR(mn, m); 503 mn->m_next = m; 504 m = mn; 505 if(m->m_flags & M_PKTHDR) { 506 if (len < MHLEN) 507 MH_ALIGN(m, len); 508 } else { 509 if (len < MLEN) 510 M_ALIGN(m, len); 511 } 512 m->m_len = len; 513 return (m); 514 } 515 516 /* 517 * Make a copy of an mbuf chain starting "off0" bytes from the beginning, 518 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf. 519 * The wait parameter is a choice of M_TRYWAIT/M_DONTWAIT from caller. 520 * Note that the copy is read-only, because clusters are not copied, 521 * only their reference counts are incremented. 522 */ 523 struct mbuf * 524 m_copym(struct mbuf *m, int off0, int len, int wait) 525 { 526 struct mbuf *n, **np; 527 int off = off0; 528 struct mbuf *top; 529 int copyhdr = 0; 530 531 KASSERT(off >= 0, ("m_copym, negative off %d", off)); 532 KASSERT(len >= 0, ("m_copym, negative len %d", len)); 533 MBUF_CHECKSLEEP(wait); 534 if (off == 0 && m->m_flags & M_PKTHDR) 535 copyhdr = 1; 536 while (off > 0) { 537 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain")); 538 if (off < m->m_len) 539 break; 540 off -= m->m_len; 541 m = m->m_next; 542 } 543 np = ⊤ 544 top = 0; 545 while (len > 0) { 546 if (m == NULL) { 547 KASSERT(len == M_COPYALL, 548 ("m_copym, length > size of mbuf chain")); 549 break; 550 } 551 if (copyhdr) 552 MGETHDR(n, wait, m->m_type); 553 else 554 MGET(n, wait, m->m_type); 555 *np = n; 556 if (n == NULL) 557 goto nospace; 558 if (copyhdr) { 559 if (!m_dup_pkthdr(n, m, wait)) 560 goto nospace; 561 if (len == M_COPYALL) 562 n->m_pkthdr.len -= off0; 563 else 564 n->m_pkthdr.len = len; 565 copyhdr = 0; 566 } 567 n->m_len = min(len, m->m_len - off); 568 if (m->m_flags & M_EXT) { 569 n->m_data = m->m_data + off; 570 mb_dupcl(n, m); 571 } else 572 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t), 573 (u_int)n->m_len); 574 if (len != M_COPYALL) 575 len -= n->m_len; 576 off = 0; 577 m = m->m_next; 578 np = &n->m_next; 579 } 580 if (top == NULL) 581 mbstat.m_mcfail++; /* XXX: No consistency. */ 582 583 return (top); 584 nospace: 585 m_freem(top); 586 mbstat.m_mcfail++; /* XXX: No consistency. */ 587 return (NULL); 588 } 589 590 /* 591 * Returns mbuf chain with new head for the prepending case. 592 * Copies from mbuf (chain) n from off for len to mbuf (chain) m 593 * either prepending or appending the data. 594 * The resulting mbuf (chain) m is fully writeable. 595 * m is destination (is made writeable) 596 * n is source, off is offset in source, len is len from offset 597 * dir, 0 append, 1 prepend 598 * how, wait or nowait 599 */ 600 601 static int 602 m_bcopyxxx(void *s, void *t, u_int len) 603 { 604 bcopy(s, t, (size_t)len); 605 return 0; 606 } 607 608 struct mbuf * 609 m_copymdata(struct mbuf *m, struct mbuf *n, int off, int len, 610 int prep, int how) 611 { 612 struct mbuf *mm, *x, *z, *prev = NULL; 613 caddr_t p; 614 int i, nlen = 0; 615 caddr_t buf[MLEN]; 616 617 KASSERT(m != NULL && n != NULL, ("m_copymdata, no target or source")); 618 KASSERT(off >= 0, ("m_copymdata, negative off %d", off)); 619 KASSERT(len >= 0, ("m_copymdata, negative len %d", len)); 620 KASSERT(prep == 0 || prep == 1, ("m_copymdata, unknown direction %d", prep)); 621 622 mm = m; 623 if (!prep) { 624 while(mm->m_next) { 625 prev = mm; 626 mm = mm->m_next; 627 } 628 } 629 for (z = n; z != NULL; z = z->m_next) 630 nlen += z->m_len; 631 if (len == M_COPYALL) 632 len = nlen - off; 633 if (off + len > nlen || len < 1) 634 return NULL; 635 636 if (!M_WRITABLE(mm)) { 637 /* XXX: Use proper m_xxx function instead. */ 638 x = m_getcl(how, MT_DATA, mm->m_flags); 639 if (x == NULL) 640 return NULL; 641 bcopy(mm->m_ext.ext_buf, x->m_ext.ext_buf, x->m_ext.ext_size); 642 p = x->m_ext.ext_buf + (mm->m_data - mm->m_ext.ext_buf); 643 x->m_data = p; 644 mm->m_next = NULL; 645 if (mm != m) 646 prev->m_next = x; 647 m_free(mm); 648 mm = x; 649 } 650 651 /* 652 * Append/prepend the data. Allocating mbufs as necessary. 653 */ 654 /* Shortcut if enough free space in first/last mbuf. */ 655 if (!prep && M_TRAILINGSPACE(mm) >= len) { 656 m_apply(n, off, len, m_bcopyxxx, mtod(mm, caddr_t) + 657 mm->m_len); 658 mm->m_len += len; 659 mm->m_pkthdr.len += len; 660 return m; 661 } 662 if (prep && M_LEADINGSPACE(mm) >= len) { 663 mm->m_data = mtod(mm, caddr_t) - len; 664 m_apply(n, off, len, m_bcopyxxx, mtod(mm, caddr_t)); 665 mm->m_len += len; 666 mm->m_pkthdr.len += len; 667 return mm; 668 } 669 670 /* Expand first/last mbuf to cluster if possible. */ 671 if (!prep && !(mm->m_flags & M_EXT) && len > M_TRAILINGSPACE(mm)) { 672 bcopy(mm->m_data, &buf, mm->m_len); 673 m_clget(mm, how); 674 if (!(mm->m_flags & M_EXT)) 675 return NULL; 676 bcopy(&buf, mm->m_ext.ext_buf, mm->m_len); 677 mm->m_data = mm->m_ext.ext_buf; 678 mm->m_pkthdr.header = NULL; 679 } 680 if (prep && !(mm->m_flags & M_EXT) && len > M_LEADINGSPACE(mm)) { 681 bcopy(mm->m_data, &buf, mm->m_len); 682 m_clget(mm, how); 683 if (!(mm->m_flags & M_EXT)) 684 return NULL; 685 bcopy(&buf, (caddr_t *)mm->m_ext.ext_buf + 686 mm->m_ext.ext_size - mm->m_len, mm->m_len); 687 mm->m_data = (caddr_t)mm->m_ext.ext_buf + 688 mm->m_ext.ext_size - mm->m_len; 689 mm->m_pkthdr.header = NULL; 690 } 691 692 /* Append/prepend as many mbuf (clusters) as necessary to fit len. */ 693 if (!prep && len > M_TRAILINGSPACE(mm)) { 694 if (!m_getm(mm, len - M_TRAILINGSPACE(mm), how, MT_DATA)) 695 return NULL; 696 } 697 if (prep && len > M_LEADINGSPACE(mm)) { 698 if (!(z = m_getm(NULL, len - M_LEADINGSPACE(mm), how, MT_DATA))) 699 return NULL; 700 i = 0; 701 for (x = z; x != NULL; x = x->m_next) { 702 i += x->m_flags & M_EXT ? x->m_ext.ext_size : 703 (x->m_flags & M_PKTHDR ? MHLEN : MLEN); 704 if (!x->m_next) 705 break; 706 } 707 z->m_data += i - len; 708 m_move_pkthdr(mm, z); 709 x->m_next = mm; 710 mm = z; 711 } 712 713 /* Seek to start position in source mbuf. Optimization for long chains. */ 714 while (off > 0) { 715 if (off < n->m_len) 716 break; 717 off -= n->m_len; 718 n = n->m_next; 719 } 720 721 /* Copy data into target mbuf. */ 722 z = mm; 723 while (len > 0) { 724 KASSERT(z != NULL, ("m_copymdata, falling off target edge")); 725 i = M_TRAILINGSPACE(z); 726 m_apply(n, off, i, m_bcopyxxx, mtod(z, caddr_t) + z->m_len); 727 z->m_len += i; 728 /* fixup pkthdr.len if necessary */ 729 if ((prep ? mm : m)->m_flags & M_PKTHDR) 730 (prep ? mm : m)->m_pkthdr.len += i; 731 off += i; 732 len -= i; 733 z = z->m_next; 734 } 735 return (prep ? mm : m); 736 } 737 738 /* 739 * Copy an entire packet, including header (which must be present). 740 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'. 741 * Note that the copy is read-only, because clusters are not copied, 742 * only their reference counts are incremented. 743 * Preserve alignment of the first mbuf so if the creator has left 744 * some room at the beginning (e.g. for inserting protocol headers) 745 * the copies still have the room available. 746 */ 747 struct mbuf * 748 m_copypacket(struct mbuf *m, int how) 749 { 750 struct mbuf *top, *n, *o; 751 752 MBUF_CHECKSLEEP(how); 753 MGET(n, how, m->m_type); 754 top = n; 755 if (n == NULL) 756 goto nospace; 757 758 if (!m_dup_pkthdr(n, m, how)) 759 goto nospace; 760 n->m_len = m->m_len; 761 if (m->m_flags & M_EXT) { 762 n->m_data = m->m_data; 763 mb_dupcl(n, m); 764 } else { 765 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat ); 766 bcopy(mtod(m, char *), mtod(n, char *), n->m_len); 767 } 768 769 m = m->m_next; 770 while (m) { 771 MGET(o, how, m->m_type); 772 if (o == NULL) 773 goto nospace; 774 775 n->m_next = o; 776 n = n->m_next; 777 778 n->m_len = m->m_len; 779 if (m->m_flags & M_EXT) { 780 n->m_data = m->m_data; 781 mb_dupcl(n, m); 782 } else { 783 bcopy(mtod(m, char *), mtod(n, char *), n->m_len); 784 } 785 786 m = m->m_next; 787 } 788 return top; 789 nospace: 790 m_freem(top); 791 mbstat.m_mcfail++; /* XXX: No consistency. */ 792 return (NULL); 793 } 794 795 /* 796 * Copy data from an mbuf chain starting "off" bytes from the beginning, 797 * continuing for "len" bytes, into the indicated buffer. 798 */ 799 void 800 m_copydata(const struct mbuf *m, int off, int len, caddr_t cp) 801 { 802 u_int count; 803 804 KASSERT(off >= 0, ("m_copydata, negative off %d", off)); 805 KASSERT(len >= 0, ("m_copydata, negative len %d", len)); 806 while (off > 0) { 807 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain")); 808 if (off < m->m_len) 809 break; 810 off -= m->m_len; 811 m = m->m_next; 812 } 813 while (len > 0) { 814 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain")); 815 count = min(m->m_len - off, len); 816 bcopy(mtod(m, caddr_t) + off, cp, count); 817 len -= count; 818 cp += count; 819 off = 0; 820 m = m->m_next; 821 } 822 } 823 824 /* 825 * Copy a packet header mbuf chain into a completely new chain, including 826 * copying any mbuf clusters. Use this instead of m_copypacket() when 827 * you need a writable copy of an mbuf chain. 828 */ 829 struct mbuf * 830 m_dup(struct mbuf *m, int how) 831 { 832 struct mbuf **p, *top = NULL; 833 int remain, moff, nsize; 834 835 MBUF_CHECKSLEEP(how); 836 /* Sanity check */ 837 if (m == NULL) 838 return (NULL); 839 M_ASSERTPKTHDR(m); 840 841 /* While there's more data, get a new mbuf, tack it on, and fill it */ 842 remain = m->m_pkthdr.len; 843 moff = 0; 844 p = ⊤ 845 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */ 846 struct mbuf *n; 847 848 /* Get the next new mbuf */ 849 if (remain >= MINCLSIZE) { 850 n = m_getcl(how, m->m_type, 0); 851 nsize = MCLBYTES; 852 } else { 853 n = m_get(how, m->m_type); 854 nsize = MLEN; 855 } 856 if (n == NULL) 857 goto nospace; 858 859 if (top == NULL) { /* First one, must be PKTHDR */ 860 if (!m_dup_pkthdr(n, m, how)) { 861 m_free(n); 862 goto nospace; 863 } 864 if ((n->m_flags & M_EXT) == 0) 865 nsize = MHLEN; 866 } 867 n->m_len = 0; 868 869 /* Link it into the new chain */ 870 *p = n; 871 p = &n->m_next; 872 873 /* Copy data from original mbuf(s) into new mbuf */ 874 while (n->m_len < nsize && m != NULL) { 875 int chunk = min(nsize - n->m_len, m->m_len - moff); 876 877 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk); 878 moff += chunk; 879 n->m_len += chunk; 880 remain -= chunk; 881 if (moff == m->m_len) { 882 m = m->m_next; 883 moff = 0; 884 } 885 } 886 887 /* Check correct total mbuf length */ 888 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL), 889 ("%s: bogus m_pkthdr.len", __func__)); 890 } 891 return (top); 892 893 nospace: 894 m_freem(top); 895 mbstat.m_mcfail++; /* XXX: No consistency. */ 896 return (NULL); 897 } 898 899 /* 900 * Concatenate mbuf chain n to m. 901 * Both chains must be of the same type (e.g. MT_DATA). 902 * Any m_pkthdr is not updated. 903 */ 904 void 905 m_cat(struct mbuf *m, struct mbuf *n) 906 { 907 while (m->m_next) 908 m = m->m_next; 909 while (n) { 910 if (m->m_flags & M_EXT || 911 m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) { 912 /* just join the two chains */ 913 m->m_next = n; 914 return; 915 } 916 /* splat the data from one into the other */ 917 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 918 (u_int)n->m_len); 919 m->m_len += n->m_len; 920 n = m_free(n); 921 } 922 } 923 924 void 925 m_adj(struct mbuf *mp, int req_len) 926 { 927 int len = req_len; 928 struct mbuf *m; 929 int count; 930 931 if ((m = mp) == NULL) 932 return; 933 if (len >= 0) { 934 /* 935 * Trim from head. 936 */ 937 while (m != NULL && len > 0) { 938 if (m->m_len <= len) { 939 len -= m->m_len; 940 m->m_len = 0; 941 m = m->m_next; 942 } else { 943 m->m_len -= len; 944 m->m_data += len; 945 len = 0; 946 } 947 } 948 m = mp; 949 if (mp->m_flags & M_PKTHDR) 950 m->m_pkthdr.len -= (req_len - len); 951 } else { 952 /* 953 * Trim from tail. Scan the mbuf chain, 954 * calculating its length and finding the last mbuf. 955 * If the adjustment only affects this mbuf, then just 956 * adjust and return. Otherwise, rescan and truncate 957 * after the remaining size. 958 */ 959 len = -len; 960 count = 0; 961 for (;;) { 962 count += m->m_len; 963 if (m->m_next == (struct mbuf *)0) 964 break; 965 m = m->m_next; 966 } 967 if (m->m_len >= len) { 968 m->m_len -= len; 969 if (mp->m_flags & M_PKTHDR) 970 mp->m_pkthdr.len -= len; 971 return; 972 } 973 count -= len; 974 if (count < 0) 975 count = 0; 976 /* 977 * Correct length for chain is "count". 978 * Find the mbuf with last data, adjust its length, 979 * and toss data from remaining mbufs on chain. 980 */ 981 m = mp; 982 if (m->m_flags & M_PKTHDR) 983 m->m_pkthdr.len = count; 984 for (; m; m = m->m_next) { 985 if (m->m_len >= count) { 986 m->m_len = count; 987 if (m->m_next != NULL) { 988 m_freem(m->m_next); 989 m->m_next = NULL; 990 } 991 break; 992 } 993 count -= m->m_len; 994 } 995 } 996 } 997 998 /* 999 * Rearange an mbuf chain so that len bytes are contiguous 1000 * and in the data area of an mbuf (so that mtod and dtom 1001 * will work for a structure of size len). Returns the resulting 1002 * mbuf chain on success, frees it and returns null on failure. 1003 * If there is room, it will add up to max_protohdr-len extra bytes to the 1004 * contiguous region in an attempt to avoid being called next time. 1005 */ 1006 struct mbuf * 1007 m_pullup(struct mbuf *n, int len) 1008 { 1009 struct mbuf *m; 1010 int count; 1011 int space; 1012 1013 /* 1014 * If first mbuf has no cluster, and has room for len bytes 1015 * without shifting current data, pullup into it, 1016 * otherwise allocate a new mbuf to prepend to the chain. 1017 */ 1018 if ((n->m_flags & M_EXT) == 0 && 1019 n->m_data + len < &n->m_dat[MLEN] && n->m_next) { 1020 if (n->m_len >= len) 1021 return (n); 1022 m = n; 1023 n = n->m_next; 1024 len -= m->m_len; 1025 } else { 1026 if (len > MHLEN) 1027 goto bad; 1028 MGET(m, M_DONTWAIT, n->m_type); 1029 if (m == NULL) 1030 goto bad; 1031 m->m_len = 0; 1032 if (n->m_flags & M_PKTHDR) 1033 M_MOVE_PKTHDR(m, n); 1034 } 1035 space = &m->m_dat[MLEN] - (m->m_data + m->m_len); 1036 do { 1037 count = min(min(max(len, max_protohdr), space), n->m_len); 1038 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 1039 (u_int)count); 1040 len -= count; 1041 m->m_len += count; 1042 n->m_len -= count; 1043 space -= count; 1044 if (n->m_len) 1045 n->m_data += count; 1046 else 1047 n = m_free(n); 1048 } while (len > 0 && n); 1049 if (len > 0) { 1050 (void) m_free(m); 1051 goto bad; 1052 } 1053 m->m_next = n; 1054 return (m); 1055 bad: 1056 m_freem(n); 1057 mbstat.m_mpfail++; /* XXX: No consistency. */ 1058 return (NULL); 1059 } 1060 1061 /* 1062 * Like m_pullup(), except a new mbuf is always allocated, and we allow 1063 * the amount of empty space before the data in the new mbuf to be specified 1064 * (in the event that the caller expects to prepend later). 1065 */ 1066 int MSFail; 1067 1068 struct mbuf * 1069 m_copyup(struct mbuf *n, int len, int dstoff) 1070 { 1071 struct mbuf *m; 1072 int count, space; 1073 1074 if (len > (MHLEN - dstoff)) 1075 goto bad; 1076 MGET(m, M_DONTWAIT, n->m_type); 1077 if (m == NULL) 1078 goto bad; 1079 m->m_len = 0; 1080 if (n->m_flags & M_PKTHDR) 1081 M_MOVE_PKTHDR(m, n); 1082 m->m_data += dstoff; 1083 space = &m->m_dat[MLEN] - (m->m_data + m->m_len); 1084 do { 1085 count = min(min(max(len, max_protohdr), space), n->m_len); 1086 memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t), 1087 (unsigned)count); 1088 len -= count; 1089 m->m_len += count; 1090 n->m_len -= count; 1091 space -= count; 1092 if (n->m_len) 1093 n->m_data += count; 1094 else 1095 n = m_free(n); 1096 } while (len > 0 && n); 1097 if (len > 0) { 1098 (void) m_free(m); 1099 goto bad; 1100 } 1101 m->m_next = n; 1102 return (m); 1103 bad: 1104 m_freem(n); 1105 MSFail++; 1106 return (NULL); 1107 } 1108 1109 /* 1110 * Partition an mbuf chain in two pieces, returning the tail -- 1111 * all but the first len0 bytes. In case of failure, it returns NULL and 1112 * attempts to restore the chain to its original state. 1113 * 1114 * Note that the resulting mbufs might be read-only, because the new 1115 * mbuf can end up sharing an mbuf cluster with the original mbuf if 1116 * the "breaking point" happens to lie within a cluster mbuf. Use the 1117 * M_WRITABLE() macro to check for this case. 1118 */ 1119 struct mbuf * 1120 m_split(struct mbuf *m0, int len0, int wait) 1121 { 1122 struct mbuf *m, *n; 1123 u_int len = len0, remain; 1124 1125 MBUF_CHECKSLEEP(wait); 1126 for (m = m0; m && len > m->m_len; m = m->m_next) 1127 len -= m->m_len; 1128 if (m == NULL) 1129 return (NULL); 1130 remain = m->m_len - len; 1131 if (m0->m_flags & M_PKTHDR) { 1132 MGETHDR(n, wait, m0->m_type); 1133 if (n == NULL) 1134 return (NULL); 1135 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif; 1136 n->m_pkthdr.len = m0->m_pkthdr.len - len0; 1137 m0->m_pkthdr.len = len0; 1138 if (m->m_flags & M_EXT) 1139 goto extpacket; 1140 if (remain > MHLEN) { 1141 /* m can't be the lead packet */ 1142 MH_ALIGN(n, 0); 1143 n->m_next = m_split(m, len, wait); 1144 if (n->m_next == NULL) { 1145 (void) m_free(n); 1146 return (NULL); 1147 } else { 1148 n->m_len = 0; 1149 return (n); 1150 } 1151 } else 1152 MH_ALIGN(n, remain); 1153 } else if (remain == 0) { 1154 n = m->m_next; 1155 m->m_next = NULL; 1156 return (n); 1157 } else { 1158 MGET(n, wait, m->m_type); 1159 if (n == NULL) 1160 return (NULL); 1161 M_ALIGN(n, remain); 1162 } 1163 extpacket: 1164 if (m->m_flags & M_EXT) { 1165 n->m_data = m->m_data + len; 1166 mb_dupcl(n, m); 1167 } else { 1168 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain); 1169 } 1170 n->m_len = remain; 1171 m->m_len = len; 1172 n->m_next = m->m_next; 1173 m->m_next = NULL; 1174 return (n); 1175 } 1176 /* 1177 * Routine to copy from device local memory into mbufs. 1178 * Note that `off' argument is offset into first mbuf of target chain from 1179 * which to begin copying the data to. 1180 */ 1181 struct mbuf * 1182 m_devget(char *buf, int totlen, int off, struct ifnet *ifp, 1183 void (*copy)(char *from, caddr_t to, u_int len)) 1184 { 1185 struct mbuf *m; 1186 struct mbuf *top = NULL, **mp = ⊤ 1187 int len; 1188 1189 if (off < 0 || off > MHLEN) 1190 return (NULL); 1191 1192 while (totlen > 0) { 1193 if (top == NULL) { /* First one, must be PKTHDR */ 1194 if (totlen + off >= MINCLSIZE) { 1195 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR); 1196 len = MCLBYTES; 1197 } else { 1198 m = m_gethdr(M_DONTWAIT, MT_DATA); 1199 len = MHLEN; 1200 1201 /* Place initial small packet/header at end of mbuf */ 1202 if (m && totlen + off + max_linkhdr <= MLEN) { 1203 m->m_data += max_linkhdr; 1204 len -= max_linkhdr; 1205 } 1206 } 1207 if (m == NULL) 1208 return NULL; 1209 m->m_pkthdr.rcvif = ifp; 1210 m->m_pkthdr.len = totlen; 1211 } else { 1212 if (totlen + off >= MINCLSIZE) { 1213 m = m_getcl(M_DONTWAIT, MT_DATA, 0); 1214 len = MCLBYTES; 1215 } else { 1216 m = m_get(M_DONTWAIT, MT_DATA); 1217 len = MLEN; 1218 } 1219 if (m == NULL) { 1220 m_freem(top); 1221 return NULL; 1222 } 1223 } 1224 if (off) { 1225 m->m_data += off; 1226 len -= off; 1227 off = 0; 1228 } 1229 m->m_len = len = min(totlen, len); 1230 if (copy) 1231 copy(buf, mtod(m, caddr_t), (u_int)len); 1232 else 1233 bcopy(buf, mtod(m, caddr_t), (u_int)len); 1234 buf += len; 1235 *mp = m; 1236 mp = &m->m_next; 1237 totlen -= len; 1238 } 1239 return (top); 1240 } 1241 1242 /* 1243 * Copy data from a buffer back into the indicated mbuf chain, 1244 * starting "off" bytes from the beginning, extending the mbuf 1245 * chain if necessary. 1246 */ 1247 void 1248 m_copyback(struct mbuf *m0, int off, int len, c_caddr_t cp) 1249 { 1250 int mlen; 1251 struct mbuf *m = m0, *n; 1252 int totlen = 0; 1253 1254 if (m0 == NULL) 1255 return; 1256 while (off > (mlen = m->m_len)) { 1257 off -= mlen; 1258 totlen += mlen; 1259 if (m->m_next == NULL) { 1260 n = m_get(M_DONTWAIT, m->m_type); 1261 if (n == NULL) 1262 goto out; 1263 bzero(mtod(n, caddr_t), MLEN); 1264 n->m_len = min(MLEN, len + off); 1265 m->m_next = n; 1266 } 1267 m = m->m_next; 1268 } 1269 while (len > 0) { 1270 mlen = min (m->m_len - off, len); 1271 bcopy(cp, off + mtod(m, caddr_t), (u_int)mlen); 1272 cp += mlen; 1273 len -= mlen; 1274 mlen += off; 1275 off = 0; 1276 totlen += mlen; 1277 if (len == 0) 1278 break; 1279 if (m->m_next == NULL) { 1280 n = m_get(M_DONTWAIT, m->m_type); 1281 if (n == NULL) 1282 break; 1283 n->m_len = min(MLEN, len); 1284 m->m_next = n; 1285 } 1286 m = m->m_next; 1287 } 1288 out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen)) 1289 m->m_pkthdr.len = totlen; 1290 } 1291 1292 /* 1293 * Append the specified data to the indicated mbuf chain, 1294 * Extend the mbuf chain if the new data does not fit in 1295 * existing space. 1296 * 1297 * Return 1 if able to complete the job; otherwise 0. 1298 */ 1299 int 1300 m_append(struct mbuf *m0, int len, c_caddr_t cp) 1301 { 1302 struct mbuf *m, *n; 1303 int remainder, space; 1304 1305 for (m = m0; m->m_next != NULL; m = m->m_next) 1306 ; 1307 remainder = len; 1308 space = M_TRAILINGSPACE(m); 1309 if (space > 0) { 1310 /* 1311 * Copy into available space. 1312 */ 1313 if (space > remainder) 1314 space = remainder; 1315 bcopy(cp, mtod(m, caddr_t) + m->m_len, space); 1316 m->m_len += space; 1317 cp += space, remainder -= space; 1318 } 1319 while (remainder > 0) { 1320 /* 1321 * Allocate a new mbuf; could check space 1322 * and allocate a cluster instead. 1323 */ 1324 n = m_get(M_DONTWAIT, m->m_type); 1325 if (n == NULL) 1326 break; 1327 n->m_len = min(MLEN, remainder); 1328 bcopy(cp, mtod(n, caddr_t), n->m_len); 1329 cp += n->m_len, remainder -= n->m_len; 1330 m->m_next = n; 1331 m = n; 1332 } 1333 if (m0->m_flags & M_PKTHDR) 1334 m0->m_pkthdr.len += len - remainder; 1335 return (remainder == 0); 1336 } 1337 1338 /* 1339 * Apply function f to the data in an mbuf chain starting "off" bytes from 1340 * the beginning, continuing for "len" bytes. 1341 */ 1342 int 1343 m_apply(struct mbuf *m, int off, int len, 1344 int (*f)(void *, void *, u_int), void *arg) 1345 { 1346 u_int count; 1347 int rval; 1348 1349 KASSERT(off >= 0, ("m_apply, negative off %d", off)); 1350 KASSERT(len >= 0, ("m_apply, negative len %d", len)); 1351 while (off > 0) { 1352 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain")); 1353 if (off < m->m_len) 1354 break; 1355 off -= m->m_len; 1356 m = m->m_next; 1357 } 1358 while (len > 0) { 1359 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain")); 1360 count = min(m->m_len - off, len); 1361 rval = (*f)(arg, mtod(m, caddr_t) + off, count); 1362 if (rval) 1363 return (rval); 1364 len -= count; 1365 off = 0; 1366 m = m->m_next; 1367 } 1368 return (0); 1369 } 1370 1371 /* 1372 * Return a pointer to mbuf/offset of location in mbuf chain. 1373 */ 1374 struct mbuf * 1375 m_getptr(struct mbuf *m, int loc, int *off) 1376 { 1377 1378 while (loc >= 0) { 1379 /* Normal end of search. */ 1380 if (m->m_len > loc) { 1381 *off = loc; 1382 return (m); 1383 } else { 1384 loc -= m->m_len; 1385 if (m->m_next == NULL) { 1386 if (loc == 0) { 1387 /* Point at the end of valid data. */ 1388 *off = m->m_len; 1389 return (m); 1390 } 1391 return (NULL); 1392 } 1393 m = m->m_next; 1394 } 1395 } 1396 return (NULL); 1397 } 1398 1399 void 1400 m_print(const struct mbuf *m, int maxlen) 1401 { 1402 int len; 1403 int pdata; 1404 const struct mbuf *m2; 1405 1406 if (m->m_flags & M_PKTHDR) 1407 len = m->m_pkthdr.len; 1408 else 1409 len = -1; 1410 m2 = m; 1411 while (m2 != NULL && (len == -1 || len)) { 1412 pdata = m2->m_len; 1413 if (maxlen != -1 && pdata > maxlen) 1414 pdata = maxlen; 1415 printf("mbuf: %p len: %d, next: %p, %b%s", m2, m2->m_len, 1416 m2->m_next, m2->m_flags, "\20\20freelist\17skipfw" 1417 "\11proto5\10proto4\7proto3\6proto2\5proto1\4rdonly" 1418 "\3eor\2pkthdr\1ext", pdata ? "" : "\n"); 1419 if (pdata) 1420 printf(", %*D\n", pdata, (u_char *)m2->m_data, "-"); 1421 if (len != -1) 1422 len -= m2->m_len; 1423 m2 = m2->m_next; 1424 } 1425 if (len > 0) 1426 printf("%d bytes unaccounted for.\n", len); 1427 return; 1428 } 1429 1430 u_int 1431 m_fixhdr(struct mbuf *m0) 1432 { 1433 u_int len; 1434 1435 len = m_length(m0, NULL); 1436 m0->m_pkthdr.len = len; 1437 return (len); 1438 } 1439 1440 u_int 1441 m_length(struct mbuf *m0, struct mbuf **last) 1442 { 1443 struct mbuf *m; 1444 u_int len; 1445 1446 len = 0; 1447 for (m = m0; m != NULL; m = m->m_next) { 1448 len += m->m_len; 1449 if (m->m_next == NULL) 1450 break; 1451 } 1452 if (last != NULL) 1453 *last = m; 1454 return (len); 1455 } 1456 1457 /* 1458 * Defragment a mbuf chain, returning the shortest possible 1459 * chain of mbufs and clusters. If allocation fails and 1460 * this cannot be completed, NULL will be returned, but 1461 * the passed in chain will be unchanged. Upon success, 1462 * the original chain will be freed, and the new chain 1463 * will be returned. 1464 * 1465 * If a non-packet header is passed in, the original 1466 * mbuf (chain?) will be returned unharmed. 1467 */ 1468 struct mbuf * 1469 m_defrag(struct mbuf *m0, int how) 1470 { 1471 struct mbuf *m_new = NULL, *m_final = NULL; 1472 int progress = 0, length; 1473 1474 MBUF_CHECKSLEEP(how); 1475 if (!(m0->m_flags & M_PKTHDR)) 1476 return (m0); 1477 1478 m_fixhdr(m0); /* Needed sanity check */ 1479 1480 #ifdef MBUF_STRESS_TEST 1481 if (m_defragrandomfailures) { 1482 int temp = arc4random() & 0xff; 1483 if (temp == 0xba) 1484 goto nospace; 1485 } 1486 #endif 1487 1488 if (m0->m_pkthdr.len > MHLEN) 1489 m_final = m_getcl(how, MT_DATA, M_PKTHDR); 1490 else 1491 m_final = m_gethdr(how, MT_DATA); 1492 1493 if (m_final == NULL) 1494 goto nospace; 1495 1496 if (m_dup_pkthdr(m_final, m0, how) == 0) 1497 goto nospace; 1498 1499 m_new = m_final; 1500 1501 while (progress < m0->m_pkthdr.len) { 1502 length = m0->m_pkthdr.len - progress; 1503 if (length > MCLBYTES) 1504 length = MCLBYTES; 1505 1506 if (m_new == NULL) { 1507 if (length > MLEN) 1508 m_new = m_getcl(how, MT_DATA, 0); 1509 else 1510 m_new = m_get(how, MT_DATA); 1511 if (m_new == NULL) 1512 goto nospace; 1513 } 1514 1515 m_copydata(m0, progress, length, mtod(m_new, caddr_t)); 1516 progress += length; 1517 m_new->m_len = length; 1518 if (m_new != m_final) 1519 m_cat(m_final, m_new); 1520 m_new = NULL; 1521 } 1522 #ifdef MBUF_STRESS_TEST 1523 if (m0->m_next == NULL) 1524 m_defraguseless++; 1525 #endif 1526 m_freem(m0); 1527 m0 = m_final; 1528 #ifdef MBUF_STRESS_TEST 1529 m_defragpackets++; 1530 m_defragbytes += m0->m_pkthdr.len; 1531 #endif 1532 return (m0); 1533 nospace: 1534 #ifdef MBUF_STRESS_TEST 1535 m_defragfailure++; 1536 #endif 1537 if (m_final) 1538 m_freem(m_final); 1539 return (NULL); 1540 } 1541 1542 #ifdef MBUF_STRESS_TEST 1543 1544 /* 1545 * Fragment an mbuf chain. There's no reason you'd ever want to do 1546 * this in normal usage, but it's great for stress testing various 1547 * mbuf consumers. 1548 * 1549 * If fragmentation is not possible, the original chain will be 1550 * returned. 1551 * 1552 * Possible length values: 1553 * 0 no fragmentation will occur 1554 * > 0 each fragment will be of the specified length 1555 * -1 each fragment will be the same random value in length 1556 * -2 each fragment's length will be entirely random 1557 * (Random values range from 1 to 256) 1558 */ 1559 struct mbuf * 1560 m_fragment(struct mbuf *m0, int how, int length) 1561 { 1562 struct mbuf *m_new = NULL, *m_final = NULL; 1563 int progress = 0; 1564 1565 if (!(m0->m_flags & M_PKTHDR)) 1566 return (m0); 1567 1568 if ((length == 0) || (length < -2)) 1569 return (m0); 1570 1571 m_fixhdr(m0); /* Needed sanity check */ 1572 1573 m_final = m_getcl(how, MT_DATA, M_PKTHDR); 1574 1575 if (m_final == NULL) 1576 goto nospace; 1577 1578 if (m_dup_pkthdr(m_final, m0, how) == 0) 1579 goto nospace; 1580 1581 m_new = m_final; 1582 1583 if (length == -1) 1584 length = 1 + (arc4random() & 255); 1585 1586 while (progress < m0->m_pkthdr.len) { 1587 int fraglen; 1588 1589 if (length > 0) 1590 fraglen = length; 1591 else 1592 fraglen = 1 + (arc4random() & 255); 1593 if (fraglen > m0->m_pkthdr.len - progress) 1594 fraglen = m0->m_pkthdr.len - progress; 1595 1596 if (fraglen > MCLBYTES) 1597 fraglen = MCLBYTES; 1598 1599 if (m_new == NULL) { 1600 m_new = m_getcl(how, MT_DATA, 0); 1601 if (m_new == NULL) 1602 goto nospace; 1603 } 1604 1605 m_copydata(m0, progress, fraglen, mtod(m_new, caddr_t)); 1606 progress += fraglen; 1607 m_new->m_len = fraglen; 1608 if (m_new != m_final) 1609 m_cat(m_final, m_new); 1610 m_new = NULL; 1611 } 1612 m_freem(m0); 1613 m0 = m_final; 1614 return (m0); 1615 nospace: 1616 if (m_final) 1617 m_freem(m_final); 1618 /* Return the original chain on failure */ 1619 return (m0); 1620 } 1621 1622 #endif 1623 1624 /* 1625 * Copy the contents of uio into a properly sized mbuf chain. 1626 */ 1627 struct mbuf * 1628 m_uiotombuf(struct uio *uio, int how, int len, int align, int flags) 1629 { 1630 struct mbuf *m, *mb; 1631 int error, length, total; 1632 int progress = 0; 1633 1634 /* 1635 * len can be zero or an arbitrary large value bound by 1636 * the total data supplied by the uio. 1637 */ 1638 if (len > 0) 1639 total = min(uio->uio_resid, len); 1640 else 1641 total = uio->uio_resid; 1642 1643 /* 1644 * The smallest unit returned by m_getm2() is a single mbuf 1645 * with pkthdr. We can't align past it. Align align itself. 1646 */ 1647 if (align) 1648 align &= ~(sizeof(long) - 1); 1649 if (align >= MHLEN) 1650 return (NULL); 1651 1652 /* 1653 * Give us the full allocation or nothing. 1654 * If len is zero return the smallest empty mbuf. 1655 */ 1656 m = m_getm2(NULL, max(total + align, 1), how, MT_DATA, flags); 1657 if (m == NULL) 1658 return (NULL); 1659 m->m_data += align; 1660 1661 /* Fill all mbufs with uio data and update header information. */ 1662 for (mb = m; mb != NULL; mb = mb->m_next) { 1663 length = min(M_TRAILINGSPACE(mb), total - progress); 1664 1665 error = uiomove(mtod(mb, void *), length, uio); 1666 if (error) { 1667 m_freem(m); 1668 return (NULL); 1669 } 1670 1671 mb->m_len = length; 1672 progress += length; 1673 if (flags & M_PKTHDR) 1674 m->m_pkthdr.len += length; 1675 } 1676 KASSERT(progress == total, ("%s: progress != total", __func__)); 1677 1678 return (m); 1679 } 1680 1681 /* 1682 * Set the m_data pointer of a newly-allocated mbuf 1683 * to place an object of the specified size at the 1684 * end of the mbuf, longword aligned. 1685 */ 1686 void 1687 m_align(struct mbuf *m, int len) 1688 { 1689 int adjust; 1690 1691 if (m->m_flags & M_EXT) 1692 adjust = m->m_ext.ext_size - len; 1693 else if (m->m_flags & M_PKTHDR) 1694 adjust = MHLEN - len; 1695 else 1696 adjust = MLEN - len; 1697 m->m_data += adjust &~ (sizeof(long)-1); 1698 } 1699 1700 /* 1701 * Create a writable copy of the mbuf chain. While doing this 1702 * we compact the chain with a goal of producing a chain with 1703 * at most two mbufs. The second mbuf in this chain is likely 1704 * to be a cluster. The primary purpose of this work is to create 1705 * a writable packet for encryption, compression, etc. The 1706 * secondary goal is to linearize the data so the data can be 1707 * passed to crypto hardware in the most efficient manner possible. 1708 */ 1709 struct mbuf * 1710 m_unshare(struct mbuf *m0, int how) 1711 { 1712 struct mbuf *m, *mprev; 1713 struct mbuf *n, *mfirst, *mlast; 1714 int len, off; 1715 1716 mprev = NULL; 1717 for (m = m0; m != NULL; m = mprev->m_next) { 1718 /* 1719 * Regular mbufs are ignored unless there's a cluster 1720 * in front of it that we can use to coalesce. We do 1721 * the latter mainly so later clusters can be coalesced 1722 * also w/o having to handle them specially (i.e. convert 1723 * mbuf+cluster -> cluster). This optimization is heavily 1724 * influenced by the assumption that we're running over 1725 * Ethernet where MCLBYTES is large enough that the max 1726 * packet size will permit lots of coalescing into a 1727 * single cluster. This in turn permits efficient 1728 * crypto operations, especially when using hardware. 1729 */ 1730 if ((m->m_flags & M_EXT) == 0) { 1731 if (mprev && (mprev->m_flags & M_EXT) && 1732 m->m_len <= M_TRAILINGSPACE(mprev)) { 1733 /* XXX: this ignores mbuf types */ 1734 memcpy(mtod(mprev, caddr_t) + mprev->m_len, 1735 mtod(m, caddr_t), m->m_len); 1736 mprev->m_len += m->m_len; 1737 mprev->m_next = m->m_next; /* unlink from chain */ 1738 m_free(m); /* reclaim mbuf */ 1739 #if 0 1740 newipsecstat.ips_mbcoalesced++; 1741 #endif 1742 } else { 1743 mprev = m; 1744 } 1745 continue; 1746 } 1747 /* 1748 * Writable mbufs are left alone (for now). 1749 */ 1750 if (M_WRITABLE(m)) { 1751 mprev = m; 1752 continue; 1753 } 1754 1755 /* 1756 * Not writable, replace with a copy or coalesce with 1757 * the previous mbuf if possible (since we have to copy 1758 * it anyway, we try to reduce the number of mbufs and 1759 * clusters so that future work is easier). 1760 */ 1761 KASSERT(m->m_flags & M_EXT, ("m_flags 0x%x", m->m_flags)); 1762 /* NB: we only coalesce into a cluster or larger */ 1763 if (mprev != NULL && (mprev->m_flags & M_EXT) && 1764 m->m_len <= M_TRAILINGSPACE(mprev)) { 1765 /* XXX: this ignores mbuf types */ 1766 memcpy(mtod(mprev, caddr_t) + mprev->m_len, 1767 mtod(m, caddr_t), m->m_len); 1768 mprev->m_len += m->m_len; 1769 mprev->m_next = m->m_next; /* unlink from chain */ 1770 m_free(m); /* reclaim mbuf */ 1771 #if 0 1772 newipsecstat.ips_clcoalesced++; 1773 #endif 1774 continue; 1775 } 1776 1777 /* 1778 * Allocate new space to hold the copy... 1779 */ 1780 /* XXX why can M_PKTHDR be set past the first mbuf? */ 1781 if (mprev == NULL && (m->m_flags & M_PKTHDR)) { 1782 /* 1783 * NB: if a packet header is present we must 1784 * allocate the mbuf separately from any cluster 1785 * because M_MOVE_PKTHDR will smash the data 1786 * pointer and drop the M_EXT marker. 1787 */ 1788 MGETHDR(n, how, m->m_type); 1789 if (n == NULL) { 1790 m_freem(m0); 1791 return (NULL); 1792 } 1793 M_MOVE_PKTHDR(n, m); 1794 MCLGET(n, how); 1795 if ((n->m_flags & M_EXT) == 0) { 1796 m_free(n); 1797 m_freem(m0); 1798 return (NULL); 1799 } 1800 } else { 1801 n = m_getcl(how, m->m_type, m->m_flags); 1802 if (n == NULL) { 1803 m_freem(m0); 1804 return (NULL); 1805 } 1806 } 1807 /* 1808 * ... and copy the data. We deal with jumbo mbufs 1809 * (i.e. m_len > MCLBYTES) by splitting them into 1810 * clusters. We could just malloc a buffer and make 1811 * it external but too many device drivers don't know 1812 * how to break up the non-contiguous memory when 1813 * doing DMA. 1814 */ 1815 len = m->m_len; 1816 off = 0; 1817 mfirst = n; 1818 mlast = NULL; 1819 for (;;) { 1820 int cc = min(len, MCLBYTES); 1821 memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + off, cc); 1822 n->m_len = cc; 1823 if (mlast != NULL) 1824 mlast->m_next = n; 1825 mlast = n; 1826 #if 0 1827 newipsecstat.ips_clcopied++; 1828 #endif 1829 1830 len -= cc; 1831 if (len <= 0) 1832 break; 1833 off += cc; 1834 1835 n = m_getcl(how, m->m_type, m->m_flags); 1836 if (n == NULL) { 1837 m_freem(mfirst); 1838 m_freem(m0); 1839 return (NULL); 1840 } 1841 } 1842 n->m_next = m->m_next; 1843 if (mprev == NULL) 1844 m0 = mfirst; /* new head of chain */ 1845 else 1846 mprev->m_next = mfirst; /* replace old mbuf */ 1847 m_free(m); /* release old mbuf */ 1848 mprev = mfirst; 1849 } 1850 return (m0); 1851 } 1852