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