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