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_param.h" 36 #include "opt_mbuf_stress_test.h" 37 #include "opt_mbuf_profiling.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 int max_linkhdr; 52 int max_protohdr; 53 int max_hdr; 54 int max_datalen; 55 #ifdef MBUF_STRESS_TEST 56 int m_defragpackets; 57 int m_defragbytes; 58 int m_defraguseless; 59 int m_defragfailure; 60 int m_defragrandomfailures; 61 #endif 62 63 /* 64 * sysctl(8) exported objects 65 */ 66 SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RD, 67 &max_linkhdr, 0, "Size of largest link layer header"); 68 SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RD, 69 &max_protohdr, 0, "Size of largest protocol layer header"); 70 SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RD, 71 &max_hdr, 0, "Size of largest link plus protocol header"); 72 SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RD, 73 &max_datalen, 0, "Minimum space left in mbuf after max_hdr"); 74 #ifdef MBUF_STRESS_TEST 75 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD, 76 &m_defragpackets, 0, ""); 77 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD, 78 &m_defragbytes, 0, ""); 79 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD, 80 &m_defraguseless, 0, ""); 81 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD, 82 &m_defragfailure, 0, ""); 83 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW, 84 &m_defragrandomfailures, 0, ""); 85 #endif 86 87 /* 88 * Ensure the correct size of various mbuf parameters. It could be off due 89 * to compiler-induced padding and alignment artifacts. 90 */ 91 CTASSERT(MSIZE - offsetof(struct mbuf, m_dat) == MLEN); 92 CTASSERT(MSIZE - offsetof(struct mbuf, m_pktdat) == MHLEN); 93 94 /* 95 * mbuf data storage should be 64-bit aligned regardless of architectural 96 * pointer size; check this is the case with and without a packet header. 97 */ 98 CTASSERT(offsetof(struct mbuf, m_dat) % 8 == 0); 99 CTASSERT(offsetof(struct mbuf, m_pktdat) % 8 == 0); 100 101 /* 102 * While the specific values here don't matter too much (i.e., +/- a few 103 * words), we do want to ensure that changes to these values are carefully 104 * reasoned about and properly documented. This is especially the case as 105 * network-protocol and device-driver modules encode these layouts, and must 106 * be recompiled if the structures change. Check these values at compile time 107 * against the ones documented in comments in mbuf.h. 108 * 109 * NB: Possibly they should be documented there via #define's and not just 110 * comments. 111 */ 112 #if defined(__LP64__) 113 CTASSERT(offsetof(struct mbuf, m_dat) == 32); 114 CTASSERT(sizeof(struct pkthdr) == 56); 115 CTASSERT(sizeof(struct m_ext) == 48); 116 #else 117 CTASSERT(offsetof(struct mbuf, m_dat) == 24); 118 CTASSERT(sizeof(struct pkthdr) == 48); 119 CTASSERT(sizeof(struct m_ext) == 28); 120 #endif 121 122 /* 123 * Assert that the queue(3) macros produce code of the same size as an old 124 * plain pointer does. 125 */ 126 #ifdef INVARIANTS 127 static struct mbuf m_assertbuf; 128 CTASSERT(sizeof(m_assertbuf.m_slist) == sizeof(m_assertbuf.m_next)); 129 CTASSERT(sizeof(m_assertbuf.m_stailq) == sizeof(m_assertbuf.m_next)); 130 CTASSERT(sizeof(m_assertbuf.m_slistpkt) == sizeof(m_assertbuf.m_nextpkt)); 131 CTASSERT(sizeof(m_assertbuf.m_stailqpkt) == sizeof(m_assertbuf.m_nextpkt)); 132 #endif 133 134 /* 135 * m_get2() allocates minimum mbuf that would fit "size" argument. 136 */ 137 struct mbuf * 138 m_get2(int size, int how, short type, int flags) 139 { 140 struct mb_args args; 141 struct mbuf *m, *n; 142 143 args.flags = flags; 144 args.type = type; 145 146 if (size <= MHLEN || (size <= MLEN && (flags & M_PKTHDR) == 0)) 147 return (uma_zalloc_arg(zone_mbuf, &args, how)); 148 if (size <= MCLBYTES) 149 return (uma_zalloc_arg(zone_pack, &args, how)); 150 151 if (size > MJUMPAGESIZE) 152 return (NULL); 153 154 m = uma_zalloc_arg(zone_mbuf, &args, how); 155 if (m == NULL) 156 return (NULL); 157 158 n = uma_zalloc_arg(zone_jumbop, m, how); 159 if (n == NULL) { 160 uma_zfree(zone_mbuf, m); 161 return (NULL); 162 } 163 164 return (m); 165 } 166 167 /* 168 * m_getjcl() returns an mbuf with a cluster of the specified size attached. 169 * For size it takes MCLBYTES, MJUMPAGESIZE, MJUM9BYTES, MJUM16BYTES. 170 */ 171 struct mbuf * 172 m_getjcl(int how, short type, int flags, int size) 173 { 174 struct mb_args args; 175 struct mbuf *m, *n; 176 uma_zone_t zone; 177 178 if (size == MCLBYTES) 179 return m_getcl(how, type, flags); 180 181 args.flags = flags; 182 args.type = type; 183 184 m = uma_zalloc_arg(zone_mbuf, &args, how); 185 if (m == NULL) 186 return (NULL); 187 188 zone = m_getzone(size); 189 n = uma_zalloc_arg(zone, m, how); 190 if (n == NULL) { 191 uma_zfree(zone_mbuf, m); 192 return (NULL); 193 } 194 return (m); 195 } 196 197 /* 198 * Allocate a given length worth of mbufs and/or clusters (whatever fits 199 * best) and return a pointer to the top of the allocated chain. If an 200 * existing mbuf chain is provided, then we will append the new chain 201 * to the existing one but still return the top of the newly allocated 202 * chain. 203 */ 204 struct mbuf * 205 m_getm2(struct mbuf *m, int len, int how, short type, int flags) 206 { 207 struct mbuf *mb, *nm = NULL, *mtail = NULL; 208 209 KASSERT(len >= 0, ("%s: len is < 0", __func__)); 210 211 /* Validate flags. */ 212 flags &= (M_PKTHDR | M_EOR); 213 214 /* Packet header mbuf must be first in chain. */ 215 if ((flags & M_PKTHDR) && m != NULL) 216 flags &= ~M_PKTHDR; 217 218 /* Loop and append maximum sized mbufs to the chain tail. */ 219 while (len > 0) { 220 if (len > MCLBYTES) 221 mb = m_getjcl(how, type, (flags & M_PKTHDR), 222 MJUMPAGESIZE); 223 else if (len >= MINCLSIZE) 224 mb = m_getcl(how, type, (flags & M_PKTHDR)); 225 else if (flags & M_PKTHDR) 226 mb = m_gethdr(how, type); 227 else 228 mb = m_get(how, type); 229 230 /* Fail the whole operation if one mbuf can't be allocated. */ 231 if (mb == NULL) { 232 if (nm != NULL) 233 m_freem(nm); 234 return (NULL); 235 } 236 237 /* Book keeping. */ 238 len -= M_SIZE(mb); 239 if (mtail != NULL) 240 mtail->m_next = mb; 241 else 242 nm = mb; 243 mtail = mb; 244 flags &= ~M_PKTHDR; /* Only valid on the first mbuf. */ 245 } 246 if (flags & M_EOR) 247 mtail->m_flags |= M_EOR; /* Only valid on the last mbuf. */ 248 249 /* If mbuf was supplied, append new chain to the end of it. */ 250 if (m != NULL) { 251 for (mtail = m; mtail->m_next != NULL; mtail = mtail->m_next) 252 ; 253 mtail->m_next = nm; 254 mtail->m_flags &= ~M_EOR; 255 } else 256 m = nm; 257 258 return (m); 259 } 260 261 /* 262 * Free an entire chain of mbufs and associated external buffers, if 263 * applicable. 264 */ 265 void 266 m_freem(struct mbuf *mb) 267 { 268 269 while (mb != NULL) 270 mb = m_free(mb); 271 } 272 273 /*- 274 * Configure a provided mbuf to refer to the provided external storage 275 * buffer and setup a reference count for said buffer. If the setting 276 * up of the reference count fails, the M_EXT bit will not be set. If 277 * successfull, the M_EXT bit is set in the mbuf's flags. 278 * 279 * Arguments: 280 * mb The existing mbuf to which to attach the provided buffer. 281 * buf The address of the provided external storage buffer. 282 * size The size of the provided buffer. 283 * freef A pointer to a routine that is responsible for freeing the 284 * provided external storage buffer. 285 * args A pointer to an argument structure (of any type) to be passed 286 * to the provided freef routine (may be NULL). 287 * flags Any other flags to be passed to the provided mbuf. 288 * type The type that the external storage buffer should be 289 * labeled with. 290 * 291 * Returns: 292 * Nothing. 293 */ 294 int 295 m_extadd(struct mbuf *mb, caddr_t buf, u_int size, 296 void (*freef)(struct mbuf *, void *, void *), void *arg1, void *arg2, 297 int flags, int type, int wait) 298 { 299 KASSERT(type != EXT_CLUSTER, ("%s: EXT_CLUSTER not allowed", __func__)); 300 301 if (type != EXT_EXTREF) 302 mb->m_ext.ext_cnt = uma_zalloc(zone_ext_refcnt, wait); 303 304 if (mb->m_ext.ext_cnt == NULL) 305 return (ENOMEM); 306 307 *(mb->m_ext.ext_cnt) = 1; 308 mb->m_flags |= (M_EXT | flags); 309 mb->m_ext.ext_buf = buf; 310 mb->m_data = mb->m_ext.ext_buf; 311 mb->m_ext.ext_size = size; 312 mb->m_ext.ext_free = freef; 313 mb->m_ext.ext_arg1 = arg1; 314 mb->m_ext.ext_arg2 = arg2; 315 mb->m_ext.ext_type = type; 316 mb->m_ext.ext_flags = 0; 317 318 return (0); 319 } 320 321 /* 322 * Non-directly-exported function to clean up after mbufs with M_EXT 323 * storage attached to them if the reference count hits 1. 324 */ 325 void 326 mb_free_ext(struct mbuf *m) 327 { 328 int freembuf; 329 330 KASSERT(m->m_flags & M_EXT, ("%s: M_EXT not set on %p", __func__, m)); 331 332 /* 333 * Check if the header is embedded in the cluster. 334 */ 335 freembuf = (m->m_flags & M_NOFREE) ? 0 : 1; 336 337 switch (m->m_ext.ext_type) { 338 case EXT_SFBUF: 339 sf_ext_free(m->m_ext.ext_arg1, m->m_ext.ext_arg2); 340 break; 341 default: 342 KASSERT(m->m_ext.ext_cnt != NULL, 343 ("%s: no refcounting pointer on %p", __func__, m)); 344 /* 345 * Free attached storage if this mbuf is the only 346 * reference to it. 347 */ 348 if (*(m->m_ext.ext_cnt) != 1) { 349 if (atomic_fetchadd_int(m->m_ext.ext_cnt, -1) != 1) 350 break; 351 } 352 353 switch (m->m_ext.ext_type) { 354 case EXT_PACKET: /* The packet zone is special. */ 355 if (*(m->m_ext.ext_cnt) == 0) 356 *(m->m_ext.ext_cnt) = 1; 357 uma_zfree(zone_pack, m); 358 return; /* Job done. */ 359 case EXT_CLUSTER: 360 uma_zfree(zone_clust, m->m_ext.ext_buf); 361 break; 362 case EXT_JUMBOP: 363 uma_zfree(zone_jumbop, m->m_ext.ext_buf); 364 break; 365 case EXT_JUMBO9: 366 uma_zfree(zone_jumbo9, m->m_ext.ext_buf); 367 break; 368 case EXT_JUMBO16: 369 uma_zfree(zone_jumbo16, m->m_ext.ext_buf); 370 break; 371 case EXT_NET_DRV: 372 case EXT_MOD_TYPE: 373 case EXT_DISPOSABLE: 374 *(m->m_ext.ext_cnt) = 0; 375 uma_zfree(zone_ext_refcnt, __DEVOLATILE(u_int *, 376 m->m_ext.ext_cnt)); 377 /* FALLTHROUGH */ 378 case EXT_EXTREF: 379 KASSERT(m->m_ext.ext_free != NULL, 380 ("%s: ext_free not set", __func__)); 381 (*(m->m_ext.ext_free))(m, m->m_ext.ext_arg1, 382 m->m_ext.ext_arg2); 383 break; 384 default: 385 KASSERT(m->m_ext.ext_type == 0, 386 ("%s: unknown ext_type", __func__)); 387 } 388 } 389 390 if (freembuf) 391 uma_zfree(zone_mbuf, m); 392 } 393 394 /* 395 * Attach the cluster from *m to *n, set up m_ext in *n 396 * and bump the refcount of the cluster. 397 */ 398 static void 399 mb_dupcl(struct mbuf *n, const struct mbuf *m) 400 { 401 402 KASSERT(m->m_flags & M_EXT, ("%s: M_EXT not set on %p", __func__, m)); 403 KASSERT(!(n->m_flags & M_EXT), ("%s: M_EXT set on %p", __func__, n)); 404 405 switch (m->m_ext.ext_type) { 406 case EXT_SFBUF: 407 sf_ext_ref(m->m_ext.ext_arg1, m->m_ext.ext_arg2); 408 break; 409 default: 410 KASSERT(m->m_ext.ext_cnt != NULL, 411 ("%s: no refcounting pointer on %p", __func__, m)); 412 if (*(m->m_ext.ext_cnt) == 1) 413 *(m->m_ext.ext_cnt) += 1; 414 else 415 atomic_add_int(m->m_ext.ext_cnt, 1); 416 } 417 418 n->m_ext = m->m_ext; 419 n->m_flags |= M_EXT; 420 n->m_flags |= m->m_flags & M_RDONLY; 421 } 422 423 void 424 m_demote_pkthdr(struct mbuf *m) 425 { 426 427 M_ASSERTPKTHDR(m); 428 429 m_tag_delete_chain(m, NULL); 430 m->m_flags &= ~M_PKTHDR; 431 bzero(&m->m_pkthdr, sizeof(struct pkthdr)); 432 } 433 434 /* 435 * Clean up mbuf (chain) from any tags and packet headers. 436 * If "all" is set then the first mbuf in the chain will be 437 * cleaned too. 438 */ 439 void 440 m_demote(struct mbuf *m0, int all, int flags) 441 { 442 struct mbuf *m; 443 444 for (m = all ? m0 : m0->m_next; m != NULL; m = m->m_next) { 445 KASSERT(m->m_nextpkt == NULL, ("%s: m_nextpkt in m %p, m0 %p", 446 __func__, m, m0)); 447 if (m->m_flags & M_PKTHDR) 448 m_demote_pkthdr(m); 449 m->m_flags = m->m_flags & (M_EXT | M_RDONLY | M_NOFREE | flags); 450 } 451 } 452 453 /* 454 * Sanity checks on mbuf (chain) for use in KASSERT() and general 455 * debugging. 456 * Returns 0 or panics when bad and 1 on all tests passed. 457 * Sanitize, 0 to run M_SANITY_ACTION, 1 to garble things so they 458 * blow up later. 459 */ 460 int 461 m_sanity(struct mbuf *m0, int sanitize) 462 { 463 struct mbuf *m; 464 caddr_t a, b; 465 int pktlen = 0; 466 467 #ifdef INVARIANTS 468 #define M_SANITY_ACTION(s) panic("mbuf %p: " s, m) 469 #else 470 #define M_SANITY_ACTION(s) printf("mbuf %p: " s, m) 471 #endif 472 473 for (m = m0; m != NULL; m = m->m_next) { 474 /* 475 * Basic pointer checks. If any of these fails then some 476 * unrelated kernel memory before or after us is trashed. 477 * No way to recover from that. 478 */ 479 a = M_START(m); 480 b = a + M_SIZE(m); 481 if ((caddr_t)m->m_data < a) 482 M_SANITY_ACTION("m_data outside mbuf data range left"); 483 if ((caddr_t)m->m_data > b) 484 M_SANITY_ACTION("m_data outside mbuf data range right"); 485 if ((caddr_t)m->m_data + m->m_len > b) 486 M_SANITY_ACTION("m_data + m_len exeeds mbuf space"); 487 488 /* m->m_nextpkt may only be set on first mbuf in chain. */ 489 if (m != m0 && m->m_nextpkt != NULL) { 490 if (sanitize) { 491 m_freem(m->m_nextpkt); 492 m->m_nextpkt = (struct mbuf *)0xDEADC0DE; 493 } else 494 M_SANITY_ACTION("m->m_nextpkt on in-chain mbuf"); 495 } 496 497 /* packet length (not mbuf length!) calculation */ 498 if (m0->m_flags & M_PKTHDR) 499 pktlen += m->m_len; 500 501 /* m_tags may only be attached to first mbuf in chain. */ 502 if (m != m0 && m->m_flags & M_PKTHDR && 503 !SLIST_EMPTY(&m->m_pkthdr.tags)) { 504 if (sanitize) { 505 m_tag_delete_chain(m, NULL); 506 /* put in 0xDEADC0DE perhaps? */ 507 } else 508 M_SANITY_ACTION("m_tags on in-chain mbuf"); 509 } 510 511 /* M_PKTHDR may only be set on first mbuf in chain */ 512 if (m != m0 && m->m_flags & M_PKTHDR) { 513 if (sanitize) { 514 bzero(&m->m_pkthdr, sizeof(m->m_pkthdr)); 515 m->m_flags &= ~M_PKTHDR; 516 /* put in 0xDEADCODE and leave hdr flag in */ 517 } else 518 M_SANITY_ACTION("M_PKTHDR on in-chain mbuf"); 519 } 520 } 521 m = m0; 522 if (pktlen && pktlen != m->m_pkthdr.len) { 523 if (sanitize) 524 m->m_pkthdr.len = 0; 525 else 526 M_SANITY_ACTION("m_pkthdr.len != mbuf chain length"); 527 } 528 return 1; 529 530 #undef M_SANITY_ACTION 531 } 532 533 534 /* 535 * "Move" mbuf pkthdr from "from" to "to". 536 * "from" must have M_PKTHDR set, and "to" must be empty. 537 */ 538 void 539 m_move_pkthdr(struct mbuf *to, struct mbuf *from) 540 { 541 542 #if 0 543 /* see below for why these are not enabled */ 544 M_ASSERTPKTHDR(to); 545 /* Note: with MAC, this may not be a good assertion. */ 546 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), 547 ("m_move_pkthdr: to has tags")); 548 #endif 549 #ifdef MAC 550 /* 551 * XXXMAC: It could be this should also occur for non-MAC? 552 */ 553 if (to->m_flags & M_PKTHDR) 554 m_tag_delete_chain(to, NULL); 555 #endif 556 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT); 557 if ((to->m_flags & M_EXT) == 0) 558 to->m_data = to->m_pktdat; 559 to->m_pkthdr = from->m_pkthdr; /* especially tags */ 560 SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */ 561 from->m_flags &= ~M_PKTHDR; 562 } 563 564 /* 565 * Duplicate "from"'s mbuf pkthdr in "to". 566 * "from" must have M_PKTHDR set, and "to" must be empty. 567 * In particular, this does a deep copy of the packet tags. 568 */ 569 int 570 m_dup_pkthdr(struct mbuf *to, const struct mbuf *from, int how) 571 { 572 573 #if 0 574 /* 575 * The mbuf allocator only initializes the pkthdr 576 * when the mbuf is allocated with m_gethdr(). Many users 577 * (e.g. m_copy*, m_prepend) use m_get() and then 578 * smash the pkthdr as needed causing these 579 * assertions to trip. For now just disable them. 580 */ 581 M_ASSERTPKTHDR(to); 582 /* Note: with MAC, this may not be a good assertion. */ 583 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), ("m_dup_pkthdr: to has tags")); 584 #endif 585 MBUF_CHECKSLEEP(how); 586 #ifdef MAC 587 if (to->m_flags & M_PKTHDR) 588 m_tag_delete_chain(to, NULL); 589 #endif 590 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT); 591 if ((to->m_flags & M_EXT) == 0) 592 to->m_data = to->m_pktdat; 593 to->m_pkthdr = from->m_pkthdr; 594 SLIST_INIT(&to->m_pkthdr.tags); 595 return (m_tag_copy_chain(to, from, how)); 596 } 597 598 /* 599 * Lesser-used path for M_PREPEND: 600 * allocate new mbuf to prepend to chain, 601 * copy junk along. 602 */ 603 struct mbuf * 604 m_prepend(struct mbuf *m, int len, int how) 605 { 606 struct mbuf *mn; 607 608 if (m->m_flags & M_PKTHDR) 609 mn = m_gethdr(how, m->m_type); 610 else 611 mn = m_get(how, m->m_type); 612 if (mn == NULL) { 613 m_freem(m); 614 return (NULL); 615 } 616 if (m->m_flags & M_PKTHDR) 617 m_move_pkthdr(mn, m); 618 mn->m_next = m; 619 m = mn; 620 if (len < M_SIZE(m)) 621 M_ALIGN(m, len); 622 m->m_len = len; 623 return (m); 624 } 625 626 /* 627 * Make a copy of an mbuf chain starting "off0" bytes from the beginning, 628 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf. 629 * The wait parameter is a choice of M_WAITOK/M_NOWAIT from caller. 630 * Note that the copy is read-only, because clusters are not copied, 631 * only their reference counts are incremented. 632 */ 633 struct mbuf * 634 m_copym(const struct mbuf *m, int off0, int len, int wait) 635 { 636 struct mbuf *n, **np; 637 int off = off0; 638 struct mbuf *top; 639 int copyhdr = 0; 640 641 KASSERT(off >= 0, ("m_copym, negative off %d", off)); 642 KASSERT(len >= 0, ("m_copym, negative len %d", len)); 643 MBUF_CHECKSLEEP(wait); 644 if (off == 0 && m->m_flags & M_PKTHDR) 645 copyhdr = 1; 646 while (off > 0) { 647 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain")); 648 if (off < m->m_len) 649 break; 650 off -= m->m_len; 651 m = m->m_next; 652 } 653 np = ⊤ 654 top = 0; 655 while (len > 0) { 656 if (m == NULL) { 657 KASSERT(len == M_COPYALL, 658 ("m_copym, length > size of mbuf chain")); 659 break; 660 } 661 if (copyhdr) 662 n = m_gethdr(wait, m->m_type); 663 else 664 n = m_get(wait, m->m_type); 665 *np = n; 666 if (n == NULL) 667 goto nospace; 668 if (copyhdr) { 669 if (!m_dup_pkthdr(n, m, wait)) 670 goto nospace; 671 if (len == M_COPYALL) 672 n->m_pkthdr.len -= off0; 673 else 674 n->m_pkthdr.len = len; 675 copyhdr = 0; 676 } 677 n->m_len = min(len, m->m_len - off); 678 if (m->m_flags & M_EXT) { 679 n->m_data = m->m_data + off; 680 mb_dupcl(n, m); 681 } else 682 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t), 683 (u_int)n->m_len); 684 if (len != M_COPYALL) 685 len -= n->m_len; 686 off = 0; 687 m = m->m_next; 688 np = &n->m_next; 689 } 690 691 return (top); 692 nospace: 693 m_freem(top); 694 return (NULL); 695 } 696 697 /* 698 * Copy an entire packet, including header (which must be present). 699 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'. 700 * Note that the copy is read-only, because clusters are not copied, 701 * only their reference counts are incremented. 702 * Preserve alignment of the first mbuf so if the creator has left 703 * some room at the beginning (e.g. for inserting protocol headers) 704 * the copies still have the room available. 705 */ 706 struct mbuf * 707 m_copypacket(struct mbuf *m, int how) 708 { 709 struct mbuf *top, *n, *o; 710 711 MBUF_CHECKSLEEP(how); 712 n = m_get(how, m->m_type); 713 top = n; 714 if (n == NULL) 715 goto nospace; 716 717 if (!m_dup_pkthdr(n, m, how)) 718 goto nospace; 719 n->m_len = m->m_len; 720 if (m->m_flags & M_EXT) { 721 n->m_data = m->m_data; 722 mb_dupcl(n, m); 723 } else { 724 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat ); 725 bcopy(mtod(m, char *), mtod(n, char *), n->m_len); 726 } 727 728 m = m->m_next; 729 while (m) { 730 o = m_get(how, m->m_type); 731 if (o == NULL) 732 goto nospace; 733 734 n->m_next = o; 735 n = n->m_next; 736 737 n->m_len = m->m_len; 738 if (m->m_flags & M_EXT) { 739 n->m_data = m->m_data; 740 mb_dupcl(n, m); 741 } else { 742 bcopy(mtod(m, char *), mtod(n, char *), n->m_len); 743 } 744 745 m = m->m_next; 746 } 747 return top; 748 nospace: 749 m_freem(top); 750 return (NULL); 751 } 752 753 /* 754 * Copy data from an mbuf chain starting "off" bytes from the beginning, 755 * continuing for "len" bytes, into the indicated buffer. 756 */ 757 void 758 m_copydata(const struct mbuf *m, int off, int len, caddr_t cp) 759 { 760 u_int count; 761 762 KASSERT(off >= 0, ("m_copydata, negative off %d", off)); 763 KASSERT(len >= 0, ("m_copydata, negative len %d", len)); 764 while (off > 0) { 765 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain")); 766 if (off < m->m_len) 767 break; 768 off -= m->m_len; 769 m = m->m_next; 770 } 771 while (len > 0) { 772 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain")); 773 count = min(m->m_len - off, len); 774 bcopy(mtod(m, caddr_t) + off, cp, count); 775 len -= count; 776 cp += count; 777 off = 0; 778 m = m->m_next; 779 } 780 } 781 782 /* 783 * Copy a packet header mbuf chain into a completely new chain, including 784 * copying any mbuf clusters. Use this instead of m_copypacket() when 785 * you need a writable copy of an mbuf chain. 786 */ 787 struct mbuf * 788 m_dup(const struct mbuf *m, int how) 789 { 790 struct mbuf **p, *top = NULL; 791 int remain, moff, nsize; 792 793 MBUF_CHECKSLEEP(how); 794 /* Sanity check */ 795 if (m == NULL) 796 return (NULL); 797 M_ASSERTPKTHDR(m); 798 799 /* While there's more data, get a new mbuf, tack it on, and fill it */ 800 remain = m->m_pkthdr.len; 801 moff = 0; 802 p = ⊤ 803 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */ 804 struct mbuf *n; 805 806 /* Get the next new mbuf */ 807 if (remain >= MINCLSIZE) { 808 n = m_getcl(how, m->m_type, 0); 809 nsize = MCLBYTES; 810 } else { 811 n = m_get(how, m->m_type); 812 nsize = MLEN; 813 } 814 if (n == NULL) 815 goto nospace; 816 817 if (top == NULL) { /* First one, must be PKTHDR */ 818 if (!m_dup_pkthdr(n, m, how)) { 819 m_free(n); 820 goto nospace; 821 } 822 if ((n->m_flags & M_EXT) == 0) 823 nsize = MHLEN; 824 n->m_flags &= ~M_RDONLY; 825 } 826 n->m_len = 0; 827 828 /* Link it into the new chain */ 829 *p = n; 830 p = &n->m_next; 831 832 /* Copy data from original mbuf(s) into new mbuf */ 833 while (n->m_len < nsize && m != NULL) { 834 int chunk = min(nsize - n->m_len, m->m_len - moff); 835 836 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk); 837 moff += chunk; 838 n->m_len += chunk; 839 remain -= chunk; 840 if (moff == m->m_len) { 841 m = m->m_next; 842 moff = 0; 843 } 844 } 845 846 /* Check correct total mbuf length */ 847 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL), 848 ("%s: bogus m_pkthdr.len", __func__)); 849 } 850 return (top); 851 852 nospace: 853 m_freem(top); 854 return (NULL); 855 } 856 857 /* 858 * Concatenate mbuf chain n to m. 859 * Both chains must be of the same type (e.g. MT_DATA). 860 * Any m_pkthdr is not updated. 861 */ 862 void 863 m_cat(struct mbuf *m, struct mbuf *n) 864 { 865 while (m->m_next) 866 m = m->m_next; 867 while (n) { 868 if (!M_WRITABLE(m) || 869 M_TRAILINGSPACE(m) < n->m_len) { 870 /* just join the two chains */ 871 m->m_next = n; 872 return; 873 } 874 /* splat the data from one into the other */ 875 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 876 (u_int)n->m_len); 877 m->m_len += n->m_len; 878 n = m_free(n); 879 } 880 } 881 882 /* 883 * Concatenate two pkthdr mbuf chains. 884 */ 885 void 886 m_catpkt(struct mbuf *m, struct mbuf *n) 887 { 888 889 M_ASSERTPKTHDR(m); 890 M_ASSERTPKTHDR(n); 891 892 m->m_pkthdr.len += n->m_pkthdr.len; 893 m_demote(n, 1, 0); 894 895 m_cat(m, n); 896 } 897 898 void 899 m_adj(struct mbuf *mp, int req_len) 900 { 901 int len = req_len; 902 struct mbuf *m; 903 int count; 904 905 if ((m = mp) == NULL) 906 return; 907 if (len >= 0) { 908 /* 909 * Trim from head. 910 */ 911 while (m != NULL && len > 0) { 912 if (m->m_len <= len) { 913 len -= m->m_len; 914 m->m_len = 0; 915 m = m->m_next; 916 } else { 917 m->m_len -= len; 918 m->m_data += len; 919 len = 0; 920 } 921 } 922 if (mp->m_flags & M_PKTHDR) 923 mp->m_pkthdr.len -= (req_len - len); 924 } else { 925 /* 926 * Trim from tail. Scan the mbuf chain, 927 * calculating its length and finding the last mbuf. 928 * If the adjustment only affects this mbuf, then just 929 * adjust and return. Otherwise, rescan and truncate 930 * after the remaining size. 931 */ 932 len = -len; 933 count = 0; 934 for (;;) { 935 count += m->m_len; 936 if (m->m_next == (struct mbuf *)0) 937 break; 938 m = m->m_next; 939 } 940 if (m->m_len >= len) { 941 m->m_len -= len; 942 if (mp->m_flags & M_PKTHDR) 943 mp->m_pkthdr.len -= len; 944 return; 945 } 946 count -= len; 947 if (count < 0) 948 count = 0; 949 /* 950 * Correct length for chain is "count". 951 * Find the mbuf with last data, adjust its length, 952 * and toss data from remaining mbufs on chain. 953 */ 954 m = mp; 955 if (m->m_flags & M_PKTHDR) 956 m->m_pkthdr.len = count; 957 for (; m; m = m->m_next) { 958 if (m->m_len >= count) { 959 m->m_len = count; 960 if (m->m_next != NULL) { 961 m_freem(m->m_next); 962 m->m_next = NULL; 963 } 964 break; 965 } 966 count -= m->m_len; 967 } 968 } 969 } 970 971 /* 972 * Rearange an mbuf chain so that len bytes are contiguous 973 * and in the data area of an mbuf (so that mtod will work 974 * for a structure of size len). Returns the resulting 975 * mbuf chain on success, frees it and returns null on failure. 976 * If there is room, it will add up to max_protohdr-len extra bytes to the 977 * contiguous region in an attempt to avoid being called next time. 978 */ 979 struct mbuf * 980 m_pullup(struct mbuf *n, int len) 981 { 982 struct mbuf *m; 983 int count; 984 int space; 985 986 /* 987 * If first mbuf has no cluster, and has room for len bytes 988 * without shifting current data, pullup into it, 989 * otherwise allocate a new mbuf to prepend to the chain. 990 */ 991 if ((n->m_flags & M_EXT) == 0 && 992 n->m_data + len < &n->m_dat[MLEN] && n->m_next) { 993 if (n->m_len >= len) 994 return (n); 995 m = n; 996 n = n->m_next; 997 len -= m->m_len; 998 } else { 999 if (len > MHLEN) 1000 goto bad; 1001 m = m_get(M_NOWAIT, n->m_type); 1002 if (m == NULL) 1003 goto bad; 1004 if (n->m_flags & M_PKTHDR) 1005 m_move_pkthdr(m, n); 1006 } 1007 space = &m->m_dat[MLEN] - (m->m_data + m->m_len); 1008 do { 1009 count = min(min(max(len, max_protohdr), space), n->m_len); 1010 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 1011 (u_int)count); 1012 len -= count; 1013 m->m_len += count; 1014 n->m_len -= count; 1015 space -= count; 1016 if (n->m_len) 1017 n->m_data += count; 1018 else 1019 n = m_free(n); 1020 } while (len > 0 && n); 1021 if (len > 0) { 1022 (void) m_free(m); 1023 goto bad; 1024 } 1025 m->m_next = n; 1026 return (m); 1027 bad: 1028 m_freem(n); 1029 return (NULL); 1030 } 1031 1032 /* 1033 * Like m_pullup(), except a new mbuf is always allocated, and we allow 1034 * the amount of empty space before the data in the new mbuf to be specified 1035 * (in the event that the caller expects to prepend later). 1036 */ 1037 struct mbuf * 1038 m_copyup(struct mbuf *n, int len, int dstoff) 1039 { 1040 struct mbuf *m; 1041 int count, space; 1042 1043 if (len > (MHLEN - dstoff)) 1044 goto bad; 1045 m = m_get(M_NOWAIT, n->m_type); 1046 if (m == NULL) 1047 goto bad; 1048 if (n->m_flags & M_PKTHDR) 1049 m_move_pkthdr(m, n); 1050 m->m_data += dstoff; 1051 space = &m->m_dat[MLEN] - (m->m_data + m->m_len); 1052 do { 1053 count = min(min(max(len, max_protohdr), space), n->m_len); 1054 memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t), 1055 (unsigned)count); 1056 len -= count; 1057 m->m_len += count; 1058 n->m_len -= count; 1059 space -= count; 1060 if (n->m_len) 1061 n->m_data += count; 1062 else 1063 n = m_free(n); 1064 } while (len > 0 && n); 1065 if (len > 0) { 1066 (void) m_free(m); 1067 goto bad; 1068 } 1069 m->m_next = n; 1070 return (m); 1071 bad: 1072 m_freem(n); 1073 return (NULL); 1074 } 1075 1076 /* 1077 * Partition an mbuf chain in two pieces, returning the tail -- 1078 * all but the first len0 bytes. In case of failure, it returns NULL and 1079 * attempts to restore the chain to its original state. 1080 * 1081 * Note that the resulting mbufs might be read-only, because the new 1082 * mbuf can end up sharing an mbuf cluster with the original mbuf if 1083 * the "breaking point" happens to lie within a cluster mbuf. Use the 1084 * M_WRITABLE() macro to check for this case. 1085 */ 1086 struct mbuf * 1087 m_split(struct mbuf *m0, int len0, int wait) 1088 { 1089 struct mbuf *m, *n; 1090 u_int len = len0, remain; 1091 1092 MBUF_CHECKSLEEP(wait); 1093 for (m = m0; m && len > m->m_len; m = m->m_next) 1094 len -= m->m_len; 1095 if (m == NULL) 1096 return (NULL); 1097 remain = m->m_len - len; 1098 if (m0->m_flags & M_PKTHDR && remain == 0) { 1099 n = m_gethdr(wait, m0->m_type); 1100 if (n == NULL) 1101 return (NULL); 1102 n->m_next = m->m_next; 1103 m->m_next = NULL; 1104 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif; 1105 n->m_pkthdr.len = m0->m_pkthdr.len - len0; 1106 m0->m_pkthdr.len = len0; 1107 return (n); 1108 } else if (m0->m_flags & M_PKTHDR) { 1109 n = m_gethdr(wait, m0->m_type); 1110 if (n == NULL) 1111 return (NULL); 1112 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif; 1113 n->m_pkthdr.len = m0->m_pkthdr.len - len0; 1114 m0->m_pkthdr.len = len0; 1115 if (m->m_flags & M_EXT) 1116 goto extpacket; 1117 if (remain > MHLEN) { 1118 /* m can't be the lead packet */ 1119 M_ALIGN(n, 0); 1120 n->m_next = m_split(m, len, wait); 1121 if (n->m_next == NULL) { 1122 (void) m_free(n); 1123 return (NULL); 1124 } else { 1125 n->m_len = 0; 1126 return (n); 1127 } 1128 } else 1129 M_ALIGN(n, remain); 1130 } else if (remain == 0) { 1131 n = m->m_next; 1132 m->m_next = NULL; 1133 return (n); 1134 } else { 1135 n = m_get(wait, m->m_type); 1136 if (n == NULL) 1137 return (NULL); 1138 M_ALIGN(n, remain); 1139 } 1140 extpacket: 1141 if (m->m_flags & M_EXT) { 1142 n->m_data = m->m_data + len; 1143 mb_dupcl(n, m); 1144 } else { 1145 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain); 1146 } 1147 n->m_len = remain; 1148 m->m_len = len; 1149 n->m_next = m->m_next; 1150 m->m_next = NULL; 1151 return (n); 1152 } 1153 /* 1154 * Routine to copy from device local memory into mbufs. 1155 * Note that `off' argument is offset into first mbuf of target chain from 1156 * which to begin copying the data to. 1157 */ 1158 struct mbuf * 1159 m_devget(char *buf, int totlen, int off, struct ifnet *ifp, 1160 void (*copy)(char *from, caddr_t to, u_int len)) 1161 { 1162 struct mbuf *m; 1163 struct mbuf *top = NULL, **mp = ⊤ 1164 int len; 1165 1166 if (off < 0 || off > MHLEN) 1167 return (NULL); 1168 1169 while (totlen > 0) { 1170 if (top == NULL) { /* First one, must be PKTHDR */ 1171 if (totlen + off >= MINCLSIZE) { 1172 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 1173 len = MCLBYTES; 1174 } else { 1175 m = m_gethdr(M_NOWAIT, MT_DATA); 1176 len = MHLEN; 1177 1178 /* Place initial small packet/header at end of mbuf */ 1179 if (m && totlen + off + max_linkhdr <= MLEN) { 1180 m->m_data += max_linkhdr; 1181 len -= max_linkhdr; 1182 } 1183 } 1184 if (m == NULL) 1185 return NULL; 1186 m->m_pkthdr.rcvif = ifp; 1187 m->m_pkthdr.len = totlen; 1188 } else { 1189 if (totlen + off >= MINCLSIZE) { 1190 m = m_getcl(M_NOWAIT, MT_DATA, 0); 1191 len = MCLBYTES; 1192 } else { 1193 m = m_get(M_NOWAIT, MT_DATA); 1194 len = MLEN; 1195 } 1196 if (m == NULL) { 1197 m_freem(top); 1198 return NULL; 1199 } 1200 } 1201 if (off) { 1202 m->m_data += off; 1203 len -= off; 1204 off = 0; 1205 } 1206 m->m_len = len = min(totlen, len); 1207 if (copy) 1208 copy(buf, mtod(m, caddr_t), (u_int)len); 1209 else 1210 bcopy(buf, mtod(m, caddr_t), (u_int)len); 1211 buf += len; 1212 *mp = m; 1213 mp = &m->m_next; 1214 totlen -= len; 1215 } 1216 return (top); 1217 } 1218 1219 /* 1220 * Copy data from a buffer back into the indicated mbuf chain, 1221 * starting "off" bytes from the beginning, extending the mbuf 1222 * chain if necessary. 1223 */ 1224 void 1225 m_copyback(struct mbuf *m0, int off, int len, c_caddr_t cp) 1226 { 1227 int mlen; 1228 struct mbuf *m = m0, *n; 1229 int totlen = 0; 1230 1231 if (m0 == NULL) 1232 return; 1233 while (off > (mlen = m->m_len)) { 1234 off -= mlen; 1235 totlen += mlen; 1236 if (m->m_next == NULL) { 1237 n = m_get(M_NOWAIT, m->m_type); 1238 if (n == NULL) 1239 goto out; 1240 bzero(mtod(n, caddr_t), MLEN); 1241 n->m_len = min(MLEN, len + off); 1242 m->m_next = n; 1243 } 1244 m = m->m_next; 1245 } 1246 while (len > 0) { 1247 if (m->m_next == NULL && (len > m->m_len - off)) { 1248 m->m_len += min(len - (m->m_len - off), 1249 M_TRAILINGSPACE(m)); 1250 } 1251 mlen = min (m->m_len - off, len); 1252 bcopy(cp, off + mtod(m, caddr_t), (u_int)mlen); 1253 cp += mlen; 1254 len -= mlen; 1255 mlen += off; 1256 off = 0; 1257 totlen += mlen; 1258 if (len == 0) 1259 break; 1260 if (m->m_next == NULL) { 1261 n = m_get(M_NOWAIT, m->m_type); 1262 if (n == NULL) 1263 break; 1264 n->m_len = min(MLEN, len); 1265 m->m_next = n; 1266 } 1267 m = m->m_next; 1268 } 1269 out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen)) 1270 m->m_pkthdr.len = totlen; 1271 } 1272 1273 /* 1274 * Append the specified data to the indicated mbuf chain, 1275 * Extend the mbuf chain if the new data does not fit in 1276 * existing space. 1277 * 1278 * Return 1 if able to complete the job; otherwise 0. 1279 */ 1280 int 1281 m_append(struct mbuf *m0, int len, c_caddr_t cp) 1282 { 1283 struct mbuf *m, *n; 1284 int remainder, space; 1285 1286 for (m = m0; m->m_next != NULL; m = m->m_next) 1287 ; 1288 remainder = len; 1289 space = M_TRAILINGSPACE(m); 1290 if (space > 0) { 1291 /* 1292 * Copy into available space. 1293 */ 1294 if (space > remainder) 1295 space = remainder; 1296 bcopy(cp, mtod(m, caddr_t) + m->m_len, space); 1297 m->m_len += space; 1298 cp += space, remainder -= space; 1299 } 1300 while (remainder > 0) { 1301 /* 1302 * Allocate a new mbuf; could check space 1303 * and allocate a cluster instead. 1304 */ 1305 n = m_get(M_NOWAIT, m->m_type); 1306 if (n == NULL) 1307 break; 1308 n->m_len = min(MLEN, remainder); 1309 bcopy(cp, mtod(n, caddr_t), n->m_len); 1310 cp += n->m_len, remainder -= n->m_len; 1311 m->m_next = n; 1312 m = n; 1313 } 1314 if (m0->m_flags & M_PKTHDR) 1315 m0->m_pkthdr.len += len - remainder; 1316 return (remainder == 0); 1317 } 1318 1319 /* 1320 * Apply function f to the data in an mbuf chain starting "off" bytes from 1321 * the beginning, continuing for "len" bytes. 1322 */ 1323 int 1324 m_apply(struct mbuf *m, int off, int len, 1325 int (*f)(void *, void *, u_int), void *arg) 1326 { 1327 u_int count; 1328 int rval; 1329 1330 KASSERT(off >= 0, ("m_apply, negative off %d", off)); 1331 KASSERT(len >= 0, ("m_apply, negative len %d", len)); 1332 while (off > 0) { 1333 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain")); 1334 if (off < m->m_len) 1335 break; 1336 off -= m->m_len; 1337 m = m->m_next; 1338 } 1339 while (len > 0) { 1340 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain")); 1341 count = min(m->m_len - off, len); 1342 rval = (*f)(arg, mtod(m, caddr_t) + off, count); 1343 if (rval) 1344 return (rval); 1345 len -= count; 1346 off = 0; 1347 m = m->m_next; 1348 } 1349 return (0); 1350 } 1351 1352 /* 1353 * Return a pointer to mbuf/offset of location in mbuf chain. 1354 */ 1355 struct mbuf * 1356 m_getptr(struct mbuf *m, int loc, int *off) 1357 { 1358 1359 while (loc >= 0) { 1360 /* Normal end of search. */ 1361 if (m->m_len > loc) { 1362 *off = loc; 1363 return (m); 1364 } else { 1365 loc -= m->m_len; 1366 if (m->m_next == NULL) { 1367 if (loc == 0) { 1368 /* Point at the end of valid data. */ 1369 *off = m->m_len; 1370 return (m); 1371 } 1372 return (NULL); 1373 } 1374 m = m->m_next; 1375 } 1376 } 1377 return (NULL); 1378 } 1379 1380 void 1381 m_print(const struct mbuf *m, int maxlen) 1382 { 1383 int len; 1384 int pdata; 1385 const struct mbuf *m2; 1386 1387 if (m == NULL) { 1388 printf("mbuf: %p\n", m); 1389 return; 1390 } 1391 1392 if (m->m_flags & M_PKTHDR) 1393 len = m->m_pkthdr.len; 1394 else 1395 len = -1; 1396 m2 = m; 1397 while (m2 != NULL && (len == -1 || len)) { 1398 pdata = m2->m_len; 1399 if (maxlen != -1 && pdata > maxlen) 1400 pdata = maxlen; 1401 printf("mbuf: %p len: %d, next: %p, %b%s", m2, m2->m_len, 1402 m2->m_next, m2->m_flags, "\20\20freelist\17skipfw" 1403 "\11proto5\10proto4\7proto3\6proto2\5proto1\4rdonly" 1404 "\3eor\2pkthdr\1ext", pdata ? "" : "\n"); 1405 if (pdata) 1406 printf(", %*D\n", pdata, (u_char *)m2->m_data, "-"); 1407 if (len != -1) 1408 len -= m2->m_len; 1409 m2 = m2->m_next; 1410 } 1411 if (len > 0) 1412 printf("%d bytes unaccounted for.\n", len); 1413 return; 1414 } 1415 1416 u_int 1417 m_fixhdr(struct mbuf *m0) 1418 { 1419 u_int len; 1420 1421 len = m_length(m0, NULL); 1422 m0->m_pkthdr.len = len; 1423 return (len); 1424 } 1425 1426 u_int 1427 m_length(struct mbuf *m0, struct mbuf **last) 1428 { 1429 struct mbuf *m; 1430 u_int len; 1431 1432 len = 0; 1433 for (m = m0; m != NULL; m = m->m_next) { 1434 len += m->m_len; 1435 if (m->m_next == NULL) 1436 break; 1437 } 1438 if (last != NULL) 1439 *last = m; 1440 return (len); 1441 } 1442 1443 /* 1444 * Defragment a mbuf chain, returning the shortest possible 1445 * chain of mbufs and clusters. If allocation fails and 1446 * this cannot be completed, NULL will be returned, but 1447 * the passed in chain will be unchanged. Upon success, 1448 * the original chain will be freed, and the new chain 1449 * will be returned. 1450 * 1451 * If a non-packet header is passed in, the original 1452 * mbuf (chain?) will be returned unharmed. 1453 */ 1454 struct mbuf * 1455 m_defrag(struct mbuf *m0, int how) 1456 { 1457 struct mbuf *m_new = NULL, *m_final = NULL; 1458 int progress = 0, length; 1459 1460 MBUF_CHECKSLEEP(how); 1461 if (!(m0->m_flags & M_PKTHDR)) 1462 return (m0); 1463 1464 m_fixhdr(m0); /* Needed sanity check */ 1465 1466 #ifdef MBUF_STRESS_TEST 1467 if (m_defragrandomfailures) { 1468 int temp = arc4random() & 0xff; 1469 if (temp == 0xba) 1470 goto nospace; 1471 } 1472 #endif 1473 1474 if (m0->m_pkthdr.len > MHLEN) 1475 m_final = m_getcl(how, MT_DATA, M_PKTHDR); 1476 else 1477 m_final = m_gethdr(how, MT_DATA); 1478 1479 if (m_final == NULL) 1480 goto nospace; 1481 1482 if (m_dup_pkthdr(m_final, m0, how) == 0) 1483 goto nospace; 1484 1485 m_new = m_final; 1486 1487 while (progress < m0->m_pkthdr.len) { 1488 length = m0->m_pkthdr.len - progress; 1489 if (length > MCLBYTES) 1490 length = MCLBYTES; 1491 1492 if (m_new == NULL) { 1493 if (length > MLEN) 1494 m_new = m_getcl(how, MT_DATA, 0); 1495 else 1496 m_new = m_get(how, MT_DATA); 1497 if (m_new == NULL) 1498 goto nospace; 1499 } 1500 1501 m_copydata(m0, progress, length, mtod(m_new, caddr_t)); 1502 progress += length; 1503 m_new->m_len = length; 1504 if (m_new != m_final) 1505 m_cat(m_final, m_new); 1506 m_new = NULL; 1507 } 1508 #ifdef MBUF_STRESS_TEST 1509 if (m0->m_next == NULL) 1510 m_defraguseless++; 1511 #endif 1512 m_freem(m0); 1513 m0 = m_final; 1514 #ifdef MBUF_STRESS_TEST 1515 m_defragpackets++; 1516 m_defragbytes += m0->m_pkthdr.len; 1517 #endif 1518 return (m0); 1519 nospace: 1520 #ifdef MBUF_STRESS_TEST 1521 m_defragfailure++; 1522 #endif 1523 if (m_final) 1524 m_freem(m_final); 1525 return (NULL); 1526 } 1527 1528 /* 1529 * Defragment an mbuf chain, returning at most maxfrags separate 1530 * mbufs+clusters. If this is not possible NULL is returned and 1531 * the original mbuf chain is left in it's present (potentially 1532 * modified) state. We use two techniques: collapsing consecutive 1533 * mbufs and replacing consecutive mbufs by a cluster. 1534 * 1535 * NB: this should really be named m_defrag but that name is taken 1536 */ 1537 struct mbuf * 1538 m_collapse(struct mbuf *m0, int how, int maxfrags) 1539 { 1540 struct mbuf *m, *n, *n2, **prev; 1541 u_int curfrags; 1542 1543 /* 1544 * Calculate the current number of frags. 1545 */ 1546 curfrags = 0; 1547 for (m = m0; m != NULL; m = m->m_next) 1548 curfrags++; 1549 /* 1550 * First, try to collapse mbufs. Note that we always collapse 1551 * towards the front so we don't need to deal with moving the 1552 * pkthdr. This may be suboptimal if the first mbuf has much 1553 * less data than the following. 1554 */ 1555 m = m0; 1556 again: 1557 for (;;) { 1558 n = m->m_next; 1559 if (n == NULL) 1560 break; 1561 if (M_WRITABLE(m) && 1562 n->m_len < M_TRAILINGSPACE(m)) { 1563 bcopy(mtod(n, void *), mtod(m, char *) + m->m_len, 1564 n->m_len); 1565 m->m_len += n->m_len; 1566 m->m_next = n->m_next; 1567 m_free(n); 1568 if (--curfrags <= maxfrags) 1569 return m0; 1570 } else 1571 m = n; 1572 } 1573 KASSERT(maxfrags > 1, 1574 ("maxfrags %u, but normal collapse failed", maxfrags)); 1575 /* 1576 * Collapse consecutive mbufs to a cluster. 1577 */ 1578 prev = &m0->m_next; /* NB: not the first mbuf */ 1579 while ((n = *prev) != NULL) { 1580 if ((n2 = n->m_next) != NULL && 1581 n->m_len + n2->m_len < MCLBYTES) { 1582 m = m_getcl(how, MT_DATA, 0); 1583 if (m == NULL) 1584 goto bad; 1585 bcopy(mtod(n, void *), mtod(m, void *), n->m_len); 1586 bcopy(mtod(n2, void *), mtod(m, char *) + n->m_len, 1587 n2->m_len); 1588 m->m_len = n->m_len + n2->m_len; 1589 m->m_next = n2->m_next; 1590 *prev = m; 1591 m_free(n); 1592 m_free(n2); 1593 if (--curfrags <= maxfrags) /* +1 cl -2 mbufs */ 1594 return m0; 1595 /* 1596 * Still not there, try the normal collapse 1597 * again before we allocate another cluster. 1598 */ 1599 goto again; 1600 } 1601 prev = &n->m_next; 1602 } 1603 /* 1604 * No place where we can collapse to a cluster; punt. 1605 * This can occur if, for example, you request 2 frags 1606 * but the packet requires that both be clusters (we 1607 * never reallocate the first mbuf to avoid moving the 1608 * packet header). 1609 */ 1610 bad: 1611 return NULL; 1612 } 1613 1614 #ifdef MBUF_STRESS_TEST 1615 1616 /* 1617 * Fragment an mbuf chain. There's no reason you'd ever want to do 1618 * this in normal usage, but it's great for stress testing various 1619 * mbuf consumers. 1620 * 1621 * If fragmentation is not possible, the original chain will be 1622 * returned. 1623 * 1624 * Possible length values: 1625 * 0 no fragmentation will occur 1626 * > 0 each fragment will be of the specified length 1627 * -1 each fragment will be the same random value in length 1628 * -2 each fragment's length will be entirely random 1629 * (Random values range from 1 to 256) 1630 */ 1631 struct mbuf * 1632 m_fragment(struct mbuf *m0, int how, int length) 1633 { 1634 struct mbuf *m_new = NULL, *m_final = NULL; 1635 int progress = 0; 1636 1637 if (!(m0->m_flags & M_PKTHDR)) 1638 return (m0); 1639 1640 if ((length == 0) || (length < -2)) 1641 return (m0); 1642 1643 m_fixhdr(m0); /* Needed sanity check */ 1644 1645 m_final = m_getcl(how, MT_DATA, M_PKTHDR); 1646 1647 if (m_final == NULL) 1648 goto nospace; 1649 1650 if (m_dup_pkthdr(m_final, m0, how) == 0) 1651 goto nospace; 1652 1653 m_new = m_final; 1654 1655 if (length == -1) 1656 length = 1 + (arc4random() & 255); 1657 1658 while (progress < m0->m_pkthdr.len) { 1659 int fraglen; 1660 1661 if (length > 0) 1662 fraglen = length; 1663 else 1664 fraglen = 1 + (arc4random() & 255); 1665 if (fraglen > m0->m_pkthdr.len - progress) 1666 fraglen = m0->m_pkthdr.len - progress; 1667 1668 if (fraglen > MCLBYTES) 1669 fraglen = MCLBYTES; 1670 1671 if (m_new == NULL) { 1672 m_new = m_getcl(how, MT_DATA, 0); 1673 if (m_new == NULL) 1674 goto nospace; 1675 } 1676 1677 m_copydata(m0, progress, fraglen, mtod(m_new, caddr_t)); 1678 progress += fraglen; 1679 m_new->m_len = fraglen; 1680 if (m_new != m_final) 1681 m_cat(m_final, m_new); 1682 m_new = NULL; 1683 } 1684 m_freem(m0); 1685 m0 = m_final; 1686 return (m0); 1687 nospace: 1688 if (m_final) 1689 m_freem(m_final); 1690 /* Return the original chain on failure */ 1691 return (m0); 1692 } 1693 1694 #endif 1695 1696 /* 1697 * Copy the contents of uio into a properly sized mbuf chain. 1698 */ 1699 struct mbuf * 1700 m_uiotombuf(struct uio *uio, int how, int len, int align, int flags) 1701 { 1702 struct mbuf *m, *mb; 1703 int error, length; 1704 ssize_t total; 1705 int progress = 0; 1706 1707 /* 1708 * len can be zero or an arbitrary large value bound by 1709 * the total data supplied by the uio. 1710 */ 1711 if (len > 0) 1712 total = min(uio->uio_resid, len); 1713 else 1714 total = uio->uio_resid; 1715 1716 /* 1717 * The smallest unit returned by m_getm2() is a single mbuf 1718 * with pkthdr. We can't align past it. 1719 */ 1720 if (align >= MHLEN) 1721 return (NULL); 1722 1723 /* 1724 * Give us the full allocation or nothing. 1725 * If len is zero return the smallest empty mbuf. 1726 */ 1727 m = m_getm2(NULL, max(total + align, 1), how, MT_DATA, flags); 1728 if (m == NULL) 1729 return (NULL); 1730 m->m_data += align; 1731 1732 /* Fill all mbufs with uio data and update header information. */ 1733 for (mb = m; mb != NULL; mb = mb->m_next) { 1734 length = min(M_TRAILINGSPACE(mb), total - progress); 1735 1736 error = uiomove(mtod(mb, void *), length, uio); 1737 if (error) { 1738 m_freem(m); 1739 return (NULL); 1740 } 1741 1742 mb->m_len = length; 1743 progress += length; 1744 if (flags & M_PKTHDR) 1745 m->m_pkthdr.len += length; 1746 } 1747 KASSERT(progress == total, ("%s: progress != total", __func__)); 1748 1749 return (m); 1750 } 1751 1752 /* 1753 * Copy an mbuf chain into a uio limited by len if set. 1754 */ 1755 int 1756 m_mbuftouio(struct uio *uio, struct mbuf *m, int len) 1757 { 1758 int error, length, total; 1759 int progress = 0; 1760 1761 if (len > 0) 1762 total = min(uio->uio_resid, len); 1763 else 1764 total = uio->uio_resid; 1765 1766 /* Fill the uio with data from the mbufs. */ 1767 for (; m != NULL; m = m->m_next) { 1768 length = min(m->m_len, total - progress); 1769 1770 error = uiomove(mtod(m, void *), length, uio); 1771 if (error) 1772 return (error); 1773 1774 progress += length; 1775 } 1776 1777 return (0); 1778 } 1779 1780 /* 1781 * Create a writable copy of the mbuf chain. While doing this 1782 * we compact the chain with a goal of producing a chain with 1783 * at most two mbufs. The second mbuf in this chain is likely 1784 * to be a cluster. The primary purpose of this work is to create 1785 * a writable packet for encryption, compression, etc. The 1786 * secondary goal is to linearize the data so the data can be 1787 * passed to crypto hardware in the most efficient manner possible. 1788 */ 1789 struct mbuf * 1790 m_unshare(struct mbuf *m0, int how) 1791 { 1792 struct mbuf *m, *mprev; 1793 struct mbuf *n, *mfirst, *mlast; 1794 int len, off; 1795 1796 mprev = NULL; 1797 for (m = m0; m != NULL; m = mprev->m_next) { 1798 /* 1799 * Regular mbufs are ignored unless there's a cluster 1800 * in front of it that we can use to coalesce. We do 1801 * the latter mainly so later clusters can be coalesced 1802 * also w/o having to handle them specially (i.e. convert 1803 * mbuf+cluster -> cluster). This optimization is heavily 1804 * influenced by the assumption that we're running over 1805 * Ethernet where MCLBYTES is large enough that the max 1806 * packet size will permit lots of coalescing into a 1807 * single cluster. This in turn permits efficient 1808 * crypto operations, especially when using hardware. 1809 */ 1810 if ((m->m_flags & M_EXT) == 0) { 1811 if (mprev && (mprev->m_flags & M_EXT) && 1812 m->m_len <= M_TRAILINGSPACE(mprev)) { 1813 /* XXX: this ignores mbuf types */ 1814 memcpy(mtod(mprev, caddr_t) + mprev->m_len, 1815 mtod(m, caddr_t), m->m_len); 1816 mprev->m_len += m->m_len; 1817 mprev->m_next = m->m_next; /* unlink from chain */ 1818 m_free(m); /* reclaim mbuf */ 1819 #if 0 1820 newipsecstat.ips_mbcoalesced++; 1821 #endif 1822 } else { 1823 mprev = m; 1824 } 1825 continue; 1826 } 1827 /* 1828 * Writable mbufs are left alone (for now). 1829 */ 1830 if (M_WRITABLE(m)) { 1831 mprev = m; 1832 continue; 1833 } 1834 1835 /* 1836 * Not writable, replace with a copy or coalesce with 1837 * the previous mbuf if possible (since we have to copy 1838 * it anyway, we try to reduce the number of mbufs and 1839 * clusters so that future work is easier). 1840 */ 1841 KASSERT(m->m_flags & M_EXT, ("m_flags 0x%x", m->m_flags)); 1842 /* NB: we only coalesce into a cluster or larger */ 1843 if (mprev != NULL && (mprev->m_flags & M_EXT) && 1844 m->m_len <= M_TRAILINGSPACE(mprev)) { 1845 /* XXX: this ignores mbuf types */ 1846 memcpy(mtod(mprev, caddr_t) + mprev->m_len, 1847 mtod(m, caddr_t), m->m_len); 1848 mprev->m_len += m->m_len; 1849 mprev->m_next = m->m_next; /* unlink from chain */ 1850 m_free(m); /* reclaim mbuf */ 1851 #if 0 1852 newipsecstat.ips_clcoalesced++; 1853 #endif 1854 continue; 1855 } 1856 1857 /* 1858 * Allocate new space to hold the copy and copy the data. 1859 * We deal with jumbo mbufs (i.e. m_len > MCLBYTES) by 1860 * splitting them into clusters. We could just malloc a 1861 * buffer and make it external but too many device drivers 1862 * don't know how to break up the non-contiguous memory when 1863 * doing DMA. 1864 */ 1865 n = m_getcl(how, m->m_type, m->m_flags); 1866 if (n == NULL) { 1867 m_freem(m0); 1868 return (NULL); 1869 } 1870 if (m->m_flags & M_PKTHDR) { 1871 KASSERT(mprev == NULL, ("%s: m0 %p, m %p has M_PKTHDR", 1872 __func__, m0, m)); 1873 m_move_pkthdr(n, m); 1874 } 1875 len = m->m_len; 1876 off = 0; 1877 mfirst = n; 1878 mlast = NULL; 1879 for (;;) { 1880 int cc = min(len, MCLBYTES); 1881 memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + off, cc); 1882 n->m_len = cc; 1883 if (mlast != NULL) 1884 mlast->m_next = n; 1885 mlast = n; 1886 #if 0 1887 newipsecstat.ips_clcopied++; 1888 #endif 1889 1890 len -= cc; 1891 if (len <= 0) 1892 break; 1893 off += cc; 1894 1895 n = m_getcl(how, m->m_type, m->m_flags); 1896 if (n == NULL) { 1897 m_freem(mfirst); 1898 m_freem(m0); 1899 return (NULL); 1900 } 1901 } 1902 n->m_next = m->m_next; 1903 if (mprev == NULL) 1904 m0 = mfirst; /* new head of chain */ 1905 else 1906 mprev->m_next = mfirst; /* replace old mbuf */ 1907 m_free(m); /* release old mbuf */ 1908 mprev = mfirst; 1909 } 1910 return (m0); 1911 } 1912 1913 #ifdef MBUF_PROFILING 1914 1915 #define MP_BUCKETS 32 /* don't just change this as things may overflow.*/ 1916 struct mbufprofile { 1917 uintmax_t wasted[MP_BUCKETS]; 1918 uintmax_t used[MP_BUCKETS]; 1919 uintmax_t segments[MP_BUCKETS]; 1920 } mbprof; 1921 1922 #define MP_MAXDIGITS 21 /* strlen("16,000,000,000,000,000,000") == 21 */ 1923 #define MP_NUMLINES 6 1924 #define MP_NUMSPERLINE 16 1925 #define MP_EXTRABYTES 64 /* > strlen("used:\nwasted:\nsegments:\n") */ 1926 /* work out max space needed and add a bit of spare space too */ 1927 #define MP_MAXLINE ((MP_MAXDIGITS+1) * MP_NUMSPERLINE) 1928 #define MP_BUFSIZE ((MP_MAXLINE * MP_NUMLINES) + 1 + MP_EXTRABYTES) 1929 1930 char mbprofbuf[MP_BUFSIZE]; 1931 1932 void 1933 m_profile(struct mbuf *m) 1934 { 1935 int segments = 0; 1936 int used = 0; 1937 int wasted = 0; 1938 1939 while (m) { 1940 segments++; 1941 used += m->m_len; 1942 if (m->m_flags & M_EXT) { 1943 wasted += MHLEN - sizeof(m->m_ext) + 1944 m->m_ext.ext_size - m->m_len; 1945 } else { 1946 if (m->m_flags & M_PKTHDR) 1947 wasted += MHLEN - m->m_len; 1948 else 1949 wasted += MLEN - m->m_len; 1950 } 1951 m = m->m_next; 1952 } 1953 /* be paranoid.. it helps */ 1954 if (segments > MP_BUCKETS - 1) 1955 segments = MP_BUCKETS - 1; 1956 if (used > 100000) 1957 used = 100000; 1958 if (wasted > 100000) 1959 wasted = 100000; 1960 /* store in the appropriate bucket */ 1961 /* don't bother locking. if it's slightly off, so what? */ 1962 mbprof.segments[segments]++; 1963 mbprof.used[fls(used)]++; 1964 mbprof.wasted[fls(wasted)]++; 1965 } 1966 1967 static void 1968 mbprof_textify(void) 1969 { 1970 int offset; 1971 char *c; 1972 uint64_t *p; 1973 1974 p = &mbprof.wasted[0]; 1975 c = mbprofbuf; 1976 offset = snprintf(c, MP_MAXLINE + 10, 1977 "wasted:\n" 1978 "%ju %ju %ju %ju %ju %ju %ju %ju " 1979 "%ju %ju %ju %ju %ju %ju %ju %ju\n", 1980 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7], 1981 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]); 1982 #ifdef BIG_ARRAY 1983 p = &mbprof.wasted[16]; 1984 c += offset; 1985 offset = snprintf(c, MP_MAXLINE, 1986 "%ju %ju %ju %ju %ju %ju %ju %ju " 1987 "%ju %ju %ju %ju %ju %ju %ju %ju\n", 1988 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7], 1989 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]); 1990 #endif 1991 p = &mbprof.used[0]; 1992 c += offset; 1993 offset = snprintf(c, MP_MAXLINE + 10, 1994 "used:\n" 1995 "%ju %ju %ju %ju %ju %ju %ju %ju " 1996 "%ju %ju %ju %ju %ju %ju %ju %ju\n", 1997 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7], 1998 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]); 1999 #ifdef BIG_ARRAY 2000 p = &mbprof.used[16]; 2001 c += offset; 2002 offset = snprintf(c, MP_MAXLINE, 2003 "%ju %ju %ju %ju %ju %ju %ju %ju " 2004 "%ju %ju %ju %ju %ju %ju %ju %ju\n", 2005 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7], 2006 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]); 2007 #endif 2008 p = &mbprof.segments[0]; 2009 c += offset; 2010 offset = snprintf(c, MP_MAXLINE + 10, 2011 "segments:\n" 2012 "%ju %ju %ju %ju %ju %ju %ju %ju " 2013 "%ju %ju %ju %ju %ju %ju %ju %ju\n", 2014 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7], 2015 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]); 2016 #ifdef BIG_ARRAY 2017 p = &mbprof.segments[16]; 2018 c += offset; 2019 offset = snprintf(c, MP_MAXLINE, 2020 "%ju %ju %ju %ju %ju %ju %ju %ju " 2021 "%ju %ju %ju %ju %ju %ju %ju %jju", 2022 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7], 2023 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]); 2024 #endif 2025 } 2026 2027 static int 2028 mbprof_handler(SYSCTL_HANDLER_ARGS) 2029 { 2030 int error; 2031 2032 mbprof_textify(); 2033 error = SYSCTL_OUT(req, mbprofbuf, strlen(mbprofbuf) + 1); 2034 return (error); 2035 } 2036 2037 static int 2038 mbprof_clr_handler(SYSCTL_HANDLER_ARGS) 2039 { 2040 int clear, error; 2041 2042 clear = 0; 2043 error = sysctl_handle_int(oidp, &clear, 0, req); 2044 if (error || !req->newptr) 2045 return (error); 2046 2047 if (clear) { 2048 bzero(&mbprof, sizeof(mbprof)); 2049 } 2050 2051 return (error); 2052 } 2053 2054 2055 SYSCTL_PROC(_kern_ipc, OID_AUTO, mbufprofile, CTLTYPE_STRING|CTLFLAG_RD, 2056 NULL, 0, mbprof_handler, "A", "mbuf profiling statistics"); 2057 2058 SYSCTL_PROC(_kern_ipc, OID_AUTO, mbufprofileclr, CTLTYPE_INT|CTLFLAG_RW, 2059 NULL, 0, mbprof_clr_handler, "I", "clear mbuf profiling statistics"); 2060 #endif 2061 2062