uipc_mbuf.c revision 254799
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: head/sys/kern/uipc_mbuf.c 254799 2013-08-24 16:57:44Z andre $"); 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 51int max_linkhdr; 52int max_protohdr; 53int max_hdr; 54int max_datalen; 55#ifdef MBUF_STRESS_TEST 56int m_defragpackets; 57int m_defragbytes; 58int m_defraguseless; 59int m_defragfailure; 60int m_defragrandomfailures; 61#endif 62 63/* 64 * sysctl(8) exported objects 65 */ 66SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RD, 67 &max_linkhdr, 0, "Size of largest link layer header"); 68SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RD, 69 &max_protohdr, 0, "Size of largest protocol layer header"); 70SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RD, 71 &max_hdr, 0, "Size of largest link plus protocol header"); 72SYSCTL_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 75SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD, 76 &m_defragpackets, 0, ""); 77SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD, 78 &m_defragbytes, 0, ""); 79SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD, 80 &m_defraguseless, 0, ""); 81SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD, 82 &m_defragfailure, 0, ""); 83SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW, 84 &m_defragrandomfailures, 0, ""); 85#endif 86 87/* 88 * m_get2() allocates minimum mbuf that would fit "size" argument. 89 */ 90struct mbuf * 91m_get2(int size, int how, short type, int flags) 92{ 93 struct mb_args args; 94 struct mbuf *m, *n; 95 96 args.flags = flags; 97 args.type = type; 98 99 if (size <= MHLEN || (size <= MLEN && (flags & M_PKTHDR) == 0)) 100 return (uma_zalloc_arg(zone_mbuf, &args, how)); 101 if (size <= MCLBYTES) 102 return (uma_zalloc_arg(zone_pack, &args, how)); 103 104 if (size > MJUMPAGESIZE) 105 return (NULL); 106 107 m = uma_zalloc_arg(zone_mbuf, &args, how); 108 if (m == NULL) 109 return (NULL); 110 111 n = uma_zalloc_arg(zone_jumbop, m, how); 112 if (n == NULL) { 113 uma_zfree(zone_mbuf, m); 114 return (NULL); 115 } 116 117 return (m); 118} 119 120/* 121 * m_getjcl() returns an mbuf with a cluster of the specified size attached. 122 * For size it takes MCLBYTES, MJUMPAGESIZE, MJUM9BYTES, MJUM16BYTES. 123 */ 124struct mbuf * 125m_getjcl(int how, short type, int flags, int size) 126{ 127 struct mb_args args; 128 struct mbuf *m, *n; 129 uma_zone_t zone; 130 131 if (size == MCLBYTES) 132 return m_getcl(how, type, flags); 133 134 args.flags = flags; 135 args.type = type; 136 137 m = uma_zalloc_arg(zone_mbuf, &args, how); 138 if (m == NULL) 139 return (NULL); 140 141 zone = m_getzone(size); 142 n = uma_zalloc_arg(zone, m, how); 143 if (n == NULL) { 144 uma_zfree(zone_mbuf, m); 145 return (NULL); 146 } 147 return (m); 148} 149 150/* 151 * Allocate a given length worth of mbufs and/or clusters (whatever fits 152 * best) and return a pointer to the top of the allocated chain. If an 153 * existing mbuf chain is provided, then we will append the new chain 154 * to the existing one but still return the top of the newly allocated 155 * chain. 156 */ 157struct mbuf * 158m_getm2(struct mbuf *m, int len, int how, short type, int flags) 159{ 160 struct mbuf *mb, *nm = NULL, *mtail = NULL; 161 162 KASSERT(len >= 0, ("%s: len is < 0", __func__)); 163 164 /* Validate flags. */ 165 flags &= (M_PKTHDR | M_EOR); 166 167 /* Packet header mbuf must be first in chain. */ 168 if ((flags & M_PKTHDR) && m != NULL) 169 flags &= ~M_PKTHDR; 170 171 /* Loop and append maximum sized mbufs to the chain tail. */ 172 while (len > 0) { 173 if (len > MCLBYTES) 174 mb = m_getjcl(how, type, (flags & M_PKTHDR), 175 MJUMPAGESIZE); 176 else if (len >= MINCLSIZE) 177 mb = m_getcl(how, type, (flags & M_PKTHDR)); 178 else if (flags & M_PKTHDR) 179 mb = m_gethdr(how, type); 180 else 181 mb = m_get(how, type); 182 183 /* Fail the whole operation if one mbuf can't be allocated. */ 184 if (mb == NULL) { 185 if (nm != NULL) 186 m_freem(nm); 187 return (NULL); 188 } 189 190 /* Book keeping. */ 191 len -= (mb->m_flags & M_EXT) ? mb->m_ext.ext_size : 192 ((mb->m_flags & M_PKTHDR) ? MHLEN : MLEN); 193 if (mtail != NULL) 194 mtail->m_next = mb; 195 else 196 nm = mb; 197 mtail = mb; 198 flags &= ~M_PKTHDR; /* Only valid on the first mbuf. */ 199 } 200 if (flags & M_EOR) 201 mtail->m_flags |= M_EOR; /* Only valid on the last mbuf. */ 202 203 /* If mbuf was supplied, append new chain to the end of it. */ 204 if (m != NULL) { 205 for (mtail = m; mtail->m_next != NULL; mtail = mtail->m_next) 206 ; 207 mtail->m_next = nm; 208 mtail->m_flags &= ~M_EOR; 209 } else 210 m = nm; 211 212 return (m); 213} 214 215/* 216 * Free an entire chain of mbufs and associated external buffers, if 217 * applicable. 218 */ 219void 220m_freem(struct mbuf *mb) 221{ 222 223 while (mb != NULL) 224 mb = m_free(mb); 225} 226 227/*- 228 * Configure a provided mbuf to refer to the provided external storage 229 * buffer and setup a reference count for said buffer. If the setting 230 * up of the reference count fails, the M_EXT bit will not be set. If 231 * successfull, the M_EXT bit is set in the mbuf's flags. 232 * 233 * Arguments: 234 * mb The existing mbuf to which to attach the provided buffer. 235 * buf The address of the provided external storage buffer. 236 * size The size of the provided buffer. 237 * freef A pointer to a routine that is responsible for freeing the 238 * provided external storage buffer. 239 * args A pointer to an argument structure (of any type) to be passed 240 * to the provided freef routine (may be NULL). 241 * flags Any other flags to be passed to the provided mbuf. 242 * type The type that the external storage buffer should be 243 * labeled with. 244 * 245 * Returns: 246 * Nothing. 247 */ 248int 249m_extadd(struct mbuf *mb, caddr_t buf, u_int size, 250 void (*freef)(struct mbuf *, void *, void *), void *arg1, void *arg2, 251 int flags, int type, int wait) 252{ 253 KASSERT(type != EXT_CLUSTER, ("%s: EXT_CLUSTER not allowed", __func__)); 254 255 if (type != EXT_EXTREF) 256 mb->m_ext.ref_cnt = uma_zalloc(zone_ext_refcnt, wait); 257 258 if (mb->m_ext.ref_cnt == NULL) 259 return (ENOMEM); 260 261 *(mb->m_ext.ref_cnt) = 1; 262 mb->m_flags |= (M_EXT | flags); 263 mb->m_ext.ext_buf = buf; 264 mb->m_data = mb->m_ext.ext_buf; 265 mb->m_ext.ext_size = size; 266 mb->m_ext.ext_free = freef; 267 mb->m_ext.ext_arg1 = arg1; 268 mb->m_ext.ext_arg2 = arg2; 269 mb->m_ext.ext_type = type; 270 mb->m_ext.ext_flags = 0; 271 272 return (0); 273} 274 275/* 276 * Non-directly-exported function to clean up after mbufs with M_EXT 277 * storage attached to them if the reference count hits 1. 278 */ 279void 280mb_free_ext(struct mbuf *m) 281{ 282 int skipmbuf; 283 284 KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__)); 285 KASSERT(m->m_ext.ref_cnt != NULL, ("%s: ref_cnt not set", __func__)); 286 287 /* 288 * check if the header is embedded in the cluster 289 */ 290 skipmbuf = (m->m_flags & M_NOFREE); 291 292 /* Free attached storage if this mbuf is the only reference to it. */ 293 if (*(m->m_ext.ref_cnt) == 1 || 294 atomic_fetchadd_int(m->m_ext.ref_cnt, -1) == 1) { 295 switch (m->m_ext.ext_type) { 296 case EXT_PACKET: /* The packet zone is special. */ 297 if (*(m->m_ext.ref_cnt) == 0) 298 *(m->m_ext.ref_cnt) = 1; 299 uma_zfree(zone_pack, m); 300 return; /* Job done. */ 301 case EXT_CLUSTER: 302 uma_zfree(zone_clust, m->m_ext.ext_buf); 303 break; 304 case EXT_JUMBOP: 305 uma_zfree(zone_jumbop, m->m_ext.ext_buf); 306 break; 307 case EXT_JUMBO9: 308 uma_zfree(zone_jumbo9, m->m_ext.ext_buf); 309 break; 310 case EXT_JUMBO16: 311 uma_zfree(zone_jumbo16, m->m_ext.ext_buf); 312 break; 313 case EXT_SFBUF: 314 case EXT_NET_DRV: 315 case EXT_MOD_TYPE: 316 case EXT_DISPOSABLE: 317 *(m->m_ext.ref_cnt) = 0; 318 uma_zfree(zone_ext_refcnt, __DEVOLATILE(u_int *, 319 m->m_ext.ref_cnt)); 320 /* FALLTHROUGH */ 321 case EXT_EXTREF: 322 KASSERT(m->m_ext.ext_free != NULL, 323 ("%s: ext_free not set", __func__)); 324 (*(m->m_ext.ext_free))(m, m->m_ext.ext_arg1, 325 m->m_ext.ext_arg2); 326 break; 327 default: 328 KASSERT(m->m_ext.ext_type == 0, 329 ("%s: unknown ext_type", __func__)); 330 } 331 } 332 if (skipmbuf) 333 return; 334 335 /* 336 * Free this mbuf back to the mbuf zone with all m_ext 337 * information purged. 338 */ 339 m->m_ext.ext_buf = NULL; 340 m->m_ext.ext_free = NULL; 341 m->m_ext.ext_arg1 = NULL; 342 m->m_ext.ext_arg2 = NULL; 343 m->m_ext.ref_cnt = NULL; 344 m->m_ext.ext_size = 0; 345 m->m_ext.ext_type = 0; 346 m->m_ext.ext_flags = 0; 347 m->m_flags &= ~M_EXT; 348 uma_zfree(zone_mbuf, m); 349} 350 351/* 352 * Attach the cluster from *m to *n, set up m_ext in *n 353 * and bump the refcount of the cluster. 354 */ 355static void 356mb_dupcl(struct mbuf *n, struct mbuf *m) 357{ 358 KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__)); 359 KASSERT(m->m_ext.ref_cnt != NULL, ("%s: ref_cnt not set", __func__)); 360 KASSERT((n->m_flags & M_EXT) == 0, ("%s: M_EXT set", __func__)); 361 362 if (*(m->m_ext.ref_cnt) == 1) 363 *(m->m_ext.ref_cnt) += 1; 364 else 365 atomic_add_int(m->m_ext.ref_cnt, 1); 366 n->m_ext.ext_buf = m->m_ext.ext_buf; 367 n->m_ext.ext_free = m->m_ext.ext_free; 368 n->m_ext.ext_arg1 = m->m_ext.ext_arg1; 369 n->m_ext.ext_arg2 = m->m_ext.ext_arg2; 370 n->m_ext.ext_size = m->m_ext.ext_size; 371 n->m_ext.ref_cnt = m->m_ext.ref_cnt; 372 n->m_ext.ext_type = m->m_ext.ext_type; 373 n->m_ext.ext_flags = m->m_ext.ext_flags; 374 n->m_flags |= M_EXT; 375 n->m_flags |= m->m_flags & M_RDONLY; 376} 377 378/* 379 * Clean up mbuf (chain) from any tags and packet headers. 380 * If "all" is set then the first mbuf in the chain will be 381 * cleaned too. 382 */ 383void 384m_demote(struct mbuf *m0, int all) 385{ 386 struct mbuf *m; 387 388 for (m = all ? m0 : m0->m_next; m != NULL; m = m->m_next) { 389 if (m->m_flags & M_PKTHDR) { 390 m_tag_delete_chain(m, NULL); 391 m->m_flags &= ~M_PKTHDR; 392 bzero(&m->m_pkthdr, sizeof(struct pkthdr)); 393 } 394 if (m != m0 && m->m_nextpkt != NULL) { 395 KASSERT(m->m_nextpkt == NULL, 396 ("%s: m_nextpkt not NULL", __func__)); 397 m_freem(m->m_nextpkt); 398 m->m_nextpkt = NULL; 399 } 400 m->m_flags = m->m_flags & (M_EXT|M_RDONLY|M_NOFREE); 401 } 402} 403 404/* 405 * Sanity checks on mbuf (chain) for use in KASSERT() and general 406 * debugging. 407 * Returns 0 or panics when bad and 1 on all tests passed. 408 * Sanitize, 0 to run M_SANITY_ACTION, 1 to garble things so they 409 * blow up later. 410 */ 411int 412m_sanity(struct mbuf *m0, int sanitize) 413{ 414 struct mbuf *m; 415 caddr_t a, b; 416 int pktlen = 0; 417 418#ifdef INVARIANTS 419#define M_SANITY_ACTION(s) panic("mbuf %p: " s, m) 420#else 421#define M_SANITY_ACTION(s) printf("mbuf %p: " s, m) 422#endif 423 424 for (m = m0; m != NULL; m = m->m_next) { 425 /* 426 * Basic pointer checks. If any of these fails then some 427 * unrelated kernel memory before or after us is trashed. 428 * No way to recover from that. 429 */ 430 a = ((m->m_flags & M_EXT) ? m->m_ext.ext_buf : 431 ((m->m_flags & M_PKTHDR) ? (caddr_t)(&m->m_pktdat) : 432 (caddr_t)(&m->m_dat)) ); 433 b = (caddr_t)(a + (m->m_flags & M_EXT ? m->m_ext.ext_size : 434 ((m->m_flags & M_PKTHDR) ? MHLEN : MLEN))); 435 if ((caddr_t)m->m_data < a) 436 M_SANITY_ACTION("m_data outside mbuf data range left"); 437 if ((caddr_t)m->m_data > b) 438 M_SANITY_ACTION("m_data outside mbuf data range right"); 439 if ((caddr_t)m->m_data + m->m_len > b) 440 M_SANITY_ACTION("m_data + m_len exeeds mbuf space"); 441 if ((m->m_flags & M_PKTHDR) && m->m_pkthdr.header) { 442 if ((caddr_t)m->m_pkthdr.header < a || 443 (caddr_t)m->m_pkthdr.header > b) 444 M_SANITY_ACTION("m_pkthdr.header outside mbuf data range"); 445 } 446 447 /* m->m_nextpkt may only be set on first mbuf in chain. */ 448 if (m != m0 && m->m_nextpkt != NULL) { 449 if (sanitize) { 450 m_freem(m->m_nextpkt); 451 m->m_nextpkt = (struct mbuf *)0xDEADC0DE; 452 } else 453 M_SANITY_ACTION("m->m_nextpkt on in-chain mbuf"); 454 } 455 456 /* packet length (not mbuf length!) calculation */ 457 if (m0->m_flags & M_PKTHDR) 458 pktlen += m->m_len; 459 460 /* m_tags may only be attached to first mbuf in chain. */ 461 if (m != m0 && m->m_flags & M_PKTHDR && 462 !SLIST_EMPTY(&m->m_pkthdr.tags)) { 463 if (sanitize) { 464 m_tag_delete_chain(m, NULL); 465 /* put in 0xDEADC0DE perhaps? */ 466 } else 467 M_SANITY_ACTION("m_tags on in-chain mbuf"); 468 } 469 470 /* M_PKTHDR may only be set on first mbuf in chain */ 471 if (m != m0 && m->m_flags & M_PKTHDR) { 472 if (sanitize) { 473 bzero(&m->m_pkthdr, sizeof(m->m_pkthdr)); 474 m->m_flags &= ~M_PKTHDR; 475 /* put in 0xDEADCODE and leave hdr flag in */ 476 } else 477 M_SANITY_ACTION("M_PKTHDR on in-chain mbuf"); 478 } 479 } 480 m = m0; 481 if (pktlen && pktlen != m->m_pkthdr.len) { 482 if (sanitize) 483 m->m_pkthdr.len = 0; 484 else 485 M_SANITY_ACTION("m_pkthdr.len != mbuf chain length"); 486 } 487 return 1; 488 489#undef M_SANITY_ACTION 490} 491 492 493/* 494 * "Move" mbuf pkthdr from "from" to "to". 495 * "from" must have M_PKTHDR set, and "to" must be empty. 496 */ 497void 498m_move_pkthdr(struct mbuf *to, struct mbuf *from) 499{ 500 501#if 0 502 /* see below for why these are not enabled */ 503 M_ASSERTPKTHDR(to); 504 /* Note: with MAC, this may not be a good assertion. */ 505 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), 506 ("m_move_pkthdr: to has tags")); 507#endif 508#ifdef MAC 509 /* 510 * XXXMAC: It could be this should also occur for non-MAC? 511 */ 512 if (to->m_flags & M_PKTHDR) 513 m_tag_delete_chain(to, NULL); 514#endif 515 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT); 516 if ((to->m_flags & M_EXT) == 0) 517 to->m_data = to->m_pktdat; 518 to->m_pkthdr = from->m_pkthdr; /* especially tags */ 519 SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */ 520 from->m_flags &= ~M_PKTHDR; 521} 522 523/* 524 * Duplicate "from"'s mbuf pkthdr in "to". 525 * "from" must have M_PKTHDR set, and "to" must be empty. 526 * In particular, this does a deep copy of the packet tags. 527 */ 528int 529m_dup_pkthdr(struct mbuf *to, struct mbuf *from, int how) 530{ 531 532#if 0 533 /* 534 * The mbuf allocator only initializes the pkthdr 535 * when the mbuf is allocated with m_gethdr(). Many users 536 * (e.g. m_copy*, m_prepend) use m_get() and then 537 * smash the pkthdr as needed causing these 538 * assertions to trip. For now just disable them. 539 */ 540 M_ASSERTPKTHDR(to); 541 /* Note: with MAC, this may not be a good assertion. */ 542 KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), ("m_dup_pkthdr: to has tags")); 543#endif 544 MBUF_CHECKSLEEP(how); 545#ifdef MAC 546 if (to->m_flags & M_PKTHDR) 547 m_tag_delete_chain(to, NULL); 548#endif 549 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT); 550 if ((to->m_flags & M_EXT) == 0) 551 to->m_data = to->m_pktdat; 552 to->m_pkthdr = from->m_pkthdr; 553 SLIST_INIT(&to->m_pkthdr.tags); 554 return (m_tag_copy_chain(to, from, MBTOM(how))); 555} 556 557/* 558 * Lesser-used path for M_PREPEND: 559 * allocate new mbuf to prepend to chain, 560 * copy junk along. 561 */ 562struct mbuf * 563m_prepend(struct mbuf *m, int len, int how) 564{ 565 struct mbuf *mn; 566 567 if (m->m_flags & M_PKTHDR) 568 mn = m_gethdr(how, m->m_type); 569 else 570 mn = m_get(how, m->m_type); 571 if (mn == NULL) { 572 m_freem(m); 573 return (NULL); 574 } 575 if (m->m_flags & M_PKTHDR) 576 m_move_pkthdr(mn, m); 577 mn->m_next = m; 578 m = mn; 579 if(m->m_flags & M_PKTHDR) { 580 if (len < MHLEN) 581 MH_ALIGN(m, len); 582 } else { 583 if (len < MLEN) 584 M_ALIGN(m, len); 585 } 586 m->m_len = len; 587 return (m); 588} 589 590/* 591 * Make a copy of an mbuf chain starting "off0" bytes from the beginning, 592 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf. 593 * The wait parameter is a choice of M_WAITOK/M_NOWAIT from caller. 594 * Note that the copy is read-only, because clusters are not copied, 595 * only their reference counts are incremented. 596 */ 597struct mbuf * 598m_copym(struct mbuf *m, int off0, int len, int wait) 599{ 600 struct mbuf *n, **np; 601 int off = off0; 602 struct mbuf *top; 603 int copyhdr = 0; 604 605 KASSERT(off >= 0, ("m_copym, negative off %d", off)); 606 KASSERT(len >= 0, ("m_copym, negative len %d", len)); 607 MBUF_CHECKSLEEP(wait); 608 if (off == 0 && m->m_flags & M_PKTHDR) 609 copyhdr = 1; 610 while (off > 0) { 611 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain")); 612 if (off < m->m_len) 613 break; 614 off -= m->m_len; 615 m = m->m_next; 616 } 617 np = ⊤ 618 top = 0; 619 while (len > 0) { 620 if (m == NULL) { 621 KASSERT(len == M_COPYALL, 622 ("m_copym, length > size of mbuf chain")); 623 break; 624 } 625 if (copyhdr) 626 n = m_gethdr(wait, m->m_type); 627 else 628 n = m_get(wait, m->m_type); 629 *np = n; 630 if (n == NULL) 631 goto nospace; 632 if (copyhdr) { 633 if (!m_dup_pkthdr(n, m, wait)) 634 goto nospace; 635 if (len == M_COPYALL) 636 n->m_pkthdr.len -= off0; 637 else 638 n->m_pkthdr.len = len; 639 copyhdr = 0; 640 } 641 n->m_len = min(len, m->m_len - off); 642 if (m->m_flags & M_EXT) { 643 n->m_data = m->m_data + off; 644 mb_dupcl(n, m); 645 } else 646 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t), 647 (u_int)n->m_len); 648 if (len != M_COPYALL) 649 len -= n->m_len; 650 off = 0; 651 m = m->m_next; 652 np = &n->m_next; 653 } 654 655 return (top); 656nospace: 657 m_freem(top); 658 return (NULL); 659} 660 661/* 662 * Returns mbuf chain with new head for the prepending case. 663 * Copies from mbuf (chain) n from off for len to mbuf (chain) m 664 * either prepending or appending the data. 665 * The resulting mbuf (chain) m is fully writeable. 666 * m is destination (is made writeable) 667 * n is source, off is offset in source, len is len from offset 668 * dir, 0 append, 1 prepend 669 * how, wait or nowait 670 */ 671 672static int 673m_bcopyxxx(void *s, void *t, u_int len) 674{ 675 bcopy(s, t, (size_t)len); 676 return 0; 677} 678 679struct mbuf * 680m_copymdata(struct mbuf *m, struct mbuf *n, int off, int len, 681 int prep, int how) 682{ 683 struct mbuf *mm, *x, *z, *prev = NULL; 684 caddr_t p; 685 int i, nlen = 0; 686 caddr_t buf[MLEN]; 687 688 KASSERT(m != NULL && n != NULL, ("m_copymdata, no target or source")); 689 KASSERT(off >= 0, ("m_copymdata, negative off %d", off)); 690 KASSERT(len >= 0, ("m_copymdata, negative len %d", len)); 691 KASSERT(prep == 0 || prep == 1, ("m_copymdata, unknown direction %d", prep)); 692 693 mm = m; 694 if (!prep) { 695 while(mm->m_next) { 696 prev = mm; 697 mm = mm->m_next; 698 } 699 } 700 for (z = n; z != NULL; z = z->m_next) 701 nlen += z->m_len; 702 if (len == M_COPYALL) 703 len = nlen - off; 704 if (off + len > nlen || len < 1) 705 return NULL; 706 707 if (!M_WRITABLE(mm)) { 708 /* XXX: Use proper m_xxx function instead. */ 709 x = m_getcl(how, MT_DATA, mm->m_flags); 710 if (x == NULL) 711 return NULL; 712 bcopy(mm->m_ext.ext_buf, x->m_ext.ext_buf, x->m_ext.ext_size); 713 p = x->m_ext.ext_buf + (mm->m_data - mm->m_ext.ext_buf); 714 x->m_data = p; 715 mm->m_next = NULL; 716 if (mm != m) 717 prev->m_next = x; 718 m_free(mm); 719 mm = x; 720 } 721 722 /* 723 * Append/prepend the data. Allocating mbufs as necessary. 724 */ 725 /* Shortcut if enough free space in first/last mbuf. */ 726 if (!prep && M_TRAILINGSPACE(mm) >= len) { 727 m_apply(n, off, len, m_bcopyxxx, mtod(mm, caddr_t) + 728 mm->m_len); 729 mm->m_len += len; 730 mm->m_pkthdr.len += len; 731 return m; 732 } 733 if (prep && M_LEADINGSPACE(mm) >= len) { 734 mm->m_data = mtod(mm, caddr_t) - len; 735 m_apply(n, off, len, m_bcopyxxx, mtod(mm, caddr_t)); 736 mm->m_len += len; 737 mm->m_pkthdr.len += len; 738 return mm; 739 } 740 741 /* Expand first/last mbuf to cluster if possible. */ 742 if (!prep && !(mm->m_flags & M_EXT) && len > M_TRAILINGSPACE(mm)) { 743 bcopy(mm->m_data, &buf, mm->m_len); 744 m_clget(mm, how); 745 if (!(mm->m_flags & M_EXT)) 746 return NULL; 747 bcopy(&buf, mm->m_ext.ext_buf, mm->m_len); 748 mm->m_data = mm->m_ext.ext_buf; 749 mm->m_pkthdr.header = NULL; 750 } 751 if (prep && !(mm->m_flags & M_EXT) && len > M_LEADINGSPACE(mm)) { 752 bcopy(mm->m_data, &buf, mm->m_len); 753 m_clget(mm, how); 754 if (!(mm->m_flags & M_EXT)) 755 return NULL; 756 bcopy(&buf, (caddr_t *)mm->m_ext.ext_buf + 757 mm->m_ext.ext_size - mm->m_len, mm->m_len); 758 mm->m_data = (caddr_t)mm->m_ext.ext_buf + 759 mm->m_ext.ext_size - mm->m_len; 760 mm->m_pkthdr.header = NULL; 761 } 762 763 /* Append/prepend as many mbuf (clusters) as necessary to fit len. */ 764 if (!prep && len > M_TRAILINGSPACE(mm)) { 765 if (!m_getm(mm, len - M_TRAILINGSPACE(mm), how, MT_DATA)) 766 return NULL; 767 } 768 if (prep && len > M_LEADINGSPACE(mm)) { 769 if (!(z = m_getm(NULL, len - M_LEADINGSPACE(mm), how, MT_DATA))) 770 return NULL; 771 i = 0; 772 for (x = z; x != NULL; x = x->m_next) { 773 i += x->m_flags & M_EXT ? x->m_ext.ext_size : 774 (x->m_flags & M_PKTHDR ? MHLEN : MLEN); 775 if (!x->m_next) 776 break; 777 } 778 z->m_data += i - len; 779 m_move_pkthdr(mm, z); 780 x->m_next = mm; 781 mm = z; 782 } 783 784 /* Seek to start position in source mbuf. Optimization for long chains. */ 785 while (off > 0) { 786 if (off < n->m_len) 787 break; 788 off -= n->m_len; 789 n = n->m_next; 790 } 791 792 /* Copy data into target mbuf. */ 793 z = mm; 794 while (len > 0) { 795 KASSERT(z != NULL, ("m_copymdata, falling off target edge")); 796 i = M_TRAILINGSPACE(z); 797 m_apply(n, off, i, m_bcopyxxx, mtod(z, caddr_t) + z->m_len); 798 z->m_len += i; 799 /* fixup pkthdr.len if necessary */ 800 if ((prep ? mm : m)->m_flags & M_PKTHDR) 801 (prep ? mm : m)->m_pkthdr.len += i; 802 off += i; 803 len -= i; 804 z = z->m_next; 805 } 806 return (prep ? mm : m); 807} 808 809/* 810 * Copy an entire packet, including header (which must be present). 811 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'. 812 * Note that the copy is read-only, because clusters are not copied, 813 * only their reference counts are incremented. 814 * Preserve alignment of the first mbuf so if the creator has left 815 * some room at the beginning (e.g. for inserting protocol headers) 816 * the copies still have the room available. 817 */ 818struct mbuf * 819m_copypacket(struct mbuf *m, int how) 820{ 821 struct mbuf *top, *n, *o; 822 823 MBUF_CHECKSLEEP(how); 824 n = m_get(how, m->m_type); 825 top = n; 826 if (n == NULL) 827 goto nospace; 828 829 if (!m_dup_pkthdr(n, m, how)) 830 goto nospace; 831 n->m_len = m->m_len; 832 if (m->m_flags & M_EXT) { 833 n->m_data = m->m_data; 834 mb_dupcl(n, m); 835 } else { 836 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat ); 837 bcopy(mtod(m, char *), mtod(n, char *), n->m_len); 838 } 839 840 m = m->m_next; 841 while (m) { 842 o = m_get(how, m->m_type); 843 if (o == NULL) 844 goto nospace; 845 846 n->m_next = o; 847 n = n->m_next; 848 849 n->m_len = m->m_len; 850 if (m->m_flags & M_EXT) { 851 n->m_data = m->m_data; 852 mb_dupcl(n, m); 853 } else { 854 bcopy(mtod(m, char *), mtod(n, char *), n->m_len); 855 } 856 857 m = m->m_next; 858 } 859 return top; 860nospace: 861 m_freem(top); 862 return (NULL); 863} 864 865/* 866 * Copy data from an mbuf chain starting "off" bytes from the beginning, 867 * continuing for "len" bytes, into the indicated buffer. 868 */ 869void 870m_copydata(const struct mbuf *m, int off, int len, caddr_t cp) 871{ 872 u_int count; 873 874 KASSERT(off >= 0, ("m_copydata, negative off %d", off)); 875 KASSERT(len >= 0, ("m_copydata, negative len %d", len)); 876 while (off > 0) { 877 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain")); 878 if (off < m->m_len) 879 break; 880 off -= m->m_len; 881 m = m->m_next; 882 } 883 while (len > 0) { 884 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain")); 885 count = min(m->m_len - off, len); 886 bcopy(mtod(m, caddr_t) + off, cp, count); 887 len -= count; 888 cp += count; 889 off = 0; 890 m = m->m_next; 891 } 892} 893 894/* 895 * Copy a packet header mbuf chain into a completely new chain, including 896 * copying any mbuf clusters. Use this instead of m_copypacket() when 897 * you need a writable copy of an mbuf chain. 898 */ 899struct mbuf * 900m_dup(struct mbuf *m, int how) 901{ 902 struct mbuf **p, *top = NULL; 903 int remain, moff, nsize; 904 905 MBUF_CHECKSLEEP(how); 906 /* Sanity check */ 907 if (m == NULL) 908 return (NULL); 909 M_ASSERTPKTHDR(m); 910 911 /* While there's more data, get a new mbuf, tack it on, and fill it */ 912 remain = m->m_pkthdr.len; 913 moff = 0; 914 p = ⊤ 915 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */ 916 struct mbuf *n; 917 918 /* Get the next new mbuf */ 919 if (remain >= MINCLSIZE) { 920 n = m_getcl(how, m->m_type, 0); 921 nsize = MCLBYTES; 922 } else { 923 n = m_get(how, m->m_type); 924 nsize = MLEN; 925 } 926 if (n == NULL) 927 goto nospace; 928 929 if (top == NULL) { /* First one, must be PKTHDR */ 930 if (!m_dup_pkthdr(n, m, how)) { 931 m_free(n); 932 goto nospace; 933 } 934 if ((n->m_flags & M_EXT) == 0) 935 nsize = MHLEN; 936 } 937 n->m_len = 0; 938 939 /* Link it into the new chain */ 940 *p = n; 941 p = &n->m_next; 942 943 /* Copy data from original mbuf(s) into new mbuf */ 944 while (n->m_len < nsize && m != NULL) { 945 int chunk = min(nsize - n->m_len, m->m_len - moff); 946 947 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk); 948 moff += chunk; 949 n->m_len += chunk; 950 remain -= chunk; 951 if (moff == m->m_len) { 952 m = m->m_next; 953 moff = 0; 954 } 955 } 956 957 /* Check correct total mbuf length */ 958 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL), 959 ("%s: bogus m_pkthdr.len", __func__)); 960 } 961 return (top); 962 963nospace: 964 m_freem(top); 965 return (NULL); 966} 967 968/* 969 * Concatenate mbuf chain n to m. 970 * Both chains must be of the same type (e.g. MT_DATA). 971 * Any m_pkthdr is not updated. 972 */ 973void 974m_cat(struct mbuf *m, struct mbuf *n) 975{ 976 while (m->m_next) 977 m = m->m_next; 978 while (n) { 979 if (!M_WRITABLE(m) || 980 M_TRAILINGSPACE(m) < n->m_len) { 981 /* just join the two chains */ 982 m->m_next = n; 983 return; 984 } 985 /* splat the data from one into the other */ 986 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 987 (u_int)n->m_len); 988 m->m_len += n->m_len; 989 n = m_free(n); 990 } 991} 992 993void 994m_adj(struct mbuf *mp, int req_len) 995{ 996 int len = req_len; 997 struct mbuf *m; 998 int count; 999 1000 if ((m = mp) == NULL) 1001 return; 1002 if (len >= 0) { 1003 /* 1004 * Trim from head. 1005 */ 1006 while (m != NULL && len > 0) { 1007 if (m->m_len <= len) { 1008 len -= m->m_len; 1009 m->m_len = 0; 1010 m = m->m_next; 1011 } else { 1012 m->m_len -= len; 1013 m->m_data += len; 1014 len = 0; 1015 } 1016 } 1017 if (mp->m_flags & M_PKTHDR) 1018 mp->m_pkthdr.len -= (req_len - len); 1019 } else { 1020 /* 1021 * Trim from tail. Scan the mbuf chain, 1022 * calculating its length and finding the last mbuf. 1023 * If the adjustment only affects this mbuf, then just 1024 * adjust and return. Otherwise, rescan and truncate 1025 * after the remaining size. 1026 */ 1027 len = -len; 1028 count = 0; 1029 for (;;) { 1030 count += m->m_len; 1031 if (m->m_next == (struct mbuf *)0) 1032 break; 1033 m = m->m_next; 1034 } 1035 if (m->m_len >= len) { 1036 m->m_len -= len; 1037 if (mp->m_flags & M_PKTHDR) 1038 mp->m_pkthdr.len -= len; 1039 return; 1040 } 1041 count -= len; 1042 if (count < 0) 1043 count = 0; 1044 /* 1045 * Correct length for chain is "count". 1046 * Find the mbuf with last data, adjust its length, 1047 * and toss data from remaining mbufs on chain. 1048 */ 1049 m = mp; 1050 if (m->m_flags & M_PKTHDR) 1051 m->m_pkthdr.len = count; 1052 for (; m; m = m->m_next) { 1053 if (m->m_len >= count) { 1054 m->m_len = count; 1055 if (m->m_next != NULL) { 1056 m_freem(m->m_next); 1057 m->m_next = NULL; 1058 } 1059 break; 1060 } 1061 count -= m->m_len; 1062 } 1063 } 1064} 1065 1066/* 1067 * Rearange an mbuf chain so that len bytes are contiguous 1068 * and in the data area of an mbuf (so that mtod will work 1069 * for a structure of size len). Returns the resulting 1070 * mbuf chain on success, frees it and returns null on failure. 1071 * If there is room, it will add up to max_protohdr-len extra bytes to the 1072 * contiguous region in an attempt to avoid being called next time. 1073 */ 1074struct mbuf * 1075m_pullup(struct mbuf *n, int len) 1076{ 1077 struct mbuf *m; 1078 int count; 1079 int space; 1080 1081 /* 1082 * If first mbuf has no cluster, and has room for len bytes 1083 * without shifting current data, pullup into it, 1084 * otherwise allocate a new mbuf to prepend to the chain. 1085 */ 1086 if ((n->m_flags & M_EXT) == 0 && 1087 n->m_data + len < &n->m_dat[MLEN] && n->m_next) { 1088 if (n->m_len >= len) 1089 return (n); 1090 m = n; 1091 n = n->m_next; 1092 len -= m->m_len; 1093 } else { 1094 if (len > MHLEN) 1095 goto bad; 1096 m = m_get(M_NOWAIT, n->m_type); 1097 if (m == NULL) 1098 goto bad; 1099 if (n->m_flags & M_PKTHDR) 1100 m_move_pkthdr(m, n); 1101 } 1102 space = &m->m_dat[MLEN] - (m->m_data + m->m_len); 1103 do { 1104 count = min(min(max(len, max_protohdr), space), n->m_len); 1105 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 1106 (u_int)count); 1107 len -= count; 1108 m->m_len += count; 1109 n->m_len -= count; 1110 space -= count; 1111 if (n->m_len) 1112 n->m_data += count; 1113 else 1114 n = m_free(n); 1115 } while (len > 0 && n); 1116 if (len > 0) { 1117 (void) m_free(m); 1118 goto bad; 1119 } 1120 m->m_next = n; 1121 return (m); 1122bad: 1123 m_freem(n); 1124 return (NULL); 1125} 1126 1127/* 1128 * Like m_pullup(), except a new mbuf is always allocated, and we allow 1129 * the amount of empty space before the data in the new mbuf to be specified 1130 * (in the event that the caller expects to prepend later). 1131 */ 1132int MSFail; 1133 1134struct mbuf * 1135m_copyup(struct mbuf *n, int len, int dstoff) 1136{ 1137 struct mbuf *m; 1138 int count, space; 1139 1140 if (len > (MHLEN - dstoff)) 1141 goto bad; 1142 m = m_get(M_NOWAIT, n->m_type); 1143 if (m == NULL) 1144 goto bad; 1145 if (n->m_flags & M_PKTHDR) 1146 m_move_pkthdr(m, n); 1147 m->m_data += dstoff; 1148 space = &m->m_dat[MLEN] - (m->m_data + m->m_len); 1149 do { 1150 count = min(min(max(len, max_protohdr), space), n->m_len); 1151 memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t), 1152 (unsigned)count); 1153 len -= count; 1154 m->m_len += count; 1155 n->m_len -= count; 1156 space -= count; 1157 if (n->m_len) 1158 n->m_data += count; 1159 else 1160 n = m_free(n); 1161 } while (len > 0 && n); 1162 if (len > 0) { 1163 (void) m_free(m); 1164 goto bad; 1165 } 1166 m->m_next = n; 1167 return (m); 1168 bad: 1169 m_freem(n); 1170 MSFail++; 1171 return (NULL); 1172} 1173 1174/* 1175 * Partition an mbuf chain in two pieces, returning the tail -- 1176 * all but the first len0 bytes. In case of failure, it returns NULL and 1177 * attempts to restore the chain to its original state. 1178 * 1179 * Note that the resulting mbufs might be read-only, because the new 1180 * mbuf can end up sharing an mbuf cluster with the original mbuf if 1181 * the "breaking point" happens to lie within a cluster mbuf. Use the 1182 * M_WRITABLE() macro to check for this case. 1183 */ 1184struct mbuf * 1185m_split(struct mbuf *m0, int len0, int wait) 1186{ 1187 struct mbuf *m, *n; 1188 u_int len = len0, remain; 1189 1190 MBUF_CHECKSLEEP(wait); 1191 for (m = m0; m && len > m->m_len; m = m->m_next) 1192 len -= m->m_len; 1193 if (m == NULL) 1194 return (NULL); 1195 remain = m->m_len - len; 1196 if (m0->m_flags & M_PKTHDR && remain == 0) { 1197 n = m_gethdr(wait, m0->m_type); 1198 return (NULL); 1199 n->m_next = m->m_next; 1200 m->m_next = NULL; 1201 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif; 1202 n->m_pkthdr.len = m0->m_pkthdr.len - len0; 1203 m0->m_pkthdr.len = len0; 1204 return (n); 1205 } else if (m0->m_flags & M_PKTHDR) { 1206 n = m_gethdr(wait, m0->m_type); 1207 if (n == NULL) 1208 return (NULL); 1209 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif; 1210 n->m_pkthdr.len = m0->m_pkthdr.len - len0; 1211 m0->m_pkthdr.len = len0; 1212 if (m->m_flags & M_EXT) 1213 goto extpacket; 1214 if (remain > MHLEN) { 1215 /* m can't be the lead packet */ 1216 MH_ALIGN(n, 0); 1217 n->m_next = m_split(m, len, wait); 1218 if (n->m_next == NULL) { 1219 (void) m_free(n); 1220 return (NULL); 1221 } else { 1222 n->m_len = 0; 1223 return (n); 1224 } 1225 } else 1226 MH_ALIGN(n, remain); 1227 } else if (remain == 0) { 1228 n = m->m_next; 1229 m->m_next = NULL; 1230 return (n); 1231 } else { 1232 n = m_get(wait, m->m_type); 1233 if (n == NULL) 1234 return (NULL); 1235 M_ALIGN(n, remain); 1236 } 1237extpacket: 1238 if (m->m_flags & M_EXT) { 1239 n->m_data = m->m_data + len; 1240 mb_dupcl(n, m); 1241 } else { 1242 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain); 1243 } 1244 n->m_len = remain; 1245 m->m_len = len; 1246 n->m_next = m->m_next; 1247 m->m_next = NULL; 1248 return (n); 1249} 1250/* 1251 * Routine to copy from device local memory into mbufs. 1252 * Note that `off' argument is offset into first mbuf of target chain from 1253 * which to begin copying the data to. 1254 */ 1255struct mbuf * 1256m_devget(char *buf, int totlen, int off, struct ifnet *ifp, 1257 void (*copy)(char *from, caddr_t to, u_int len)) 1258{ 1259 struct mbuf *m; 1260 struct mbuf *top = NULL, **mp = ⊤ 1261 int len; 1262 1263 if (off < 0 || off > MHLEN) 1264 return (NULL); 1265 1266 while (totlen > 0) { 1267 if (top == NULL) { /* First one, must be PKTHDR */ 1268 if (totlen + off >= MINCLSIZE) { 1269 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 1270 len = MCLBYTES; 1271 } else { 1272 m = m_gethdr(M_NOWAIT, MT_DATA); 1273 len = MHLEN; 1274 1275 /* Place initial small packet/header at end of mbuf */ 1276 if (m && totlen + off + max_linkhdr <= MLEN) { 1277 m->m_data += max_linkhdr; 1278 len -= max_linkhdr; 1279 } 1280 } 1281 if (m == NULL) 1282 return NULL; 1283 m->m_pkthdr.rcvif = ifp; 1284 m->m_pkthdr.len = totlen; 1285 } else { 1286 if (totlen + off >= MINCLSIZE) { 1287 m = m_getcl(M_NOWAIT, MT_DATA, 0); 1288 len = MCLBYTES; 1289 } else { 1290 m = m_get(M_NOWAIT, MT_DATA); 1291 len = MLEN; 1292 } 1293 if (m == NULL) { 1294 m_freem(top); 1295 return NULL; 1296 } 1297 } 1298 if (off) { 1299 m->m_data += off; 1300 len -= off; 1301 off = 0; 1302 } 1303 m->m_len = len = min(totlen, len); 1304 if (copy) 1305 copy(buf, mtod(m, caddr_t), (u_int)len); 1306 else 1307 bcopy(buf, mtod(m, caddr_t), (u_int)len); 1308 buf += len; 1309 *mp = m; 1310 mp = &m->m_next; 1311 totlen -= len; 1312 } 1313 return (top); 1314} 1315 1316/* 1317 * Copy data from a buffer back into the indicated mbuf chain, 1318 * starting "off" bytes from the beginning, extending the mbuf 1319 * chain if necessary. 1320 */ 1321void 1322m_copyback(struct mbuf *m0, int off, int len, c_caddr_t cp) 1323{ 1324 int mlen; 1325 struct mbuf *m = m0, *n; 1326 int totlen = 0; 1327 1328 if (m0 == NULL) 1329 return; 1330 while (off > (mlen = m->m_len)) { 1331 off -= mlen; 1332 totlen += mlen; 1333 if (m->m_next == NULL) { 1334 n = m_get(M_NOWAIT, m->m_type); 1335 if (n == NULL) 1336 goto out; 1337 bzero(mtod(n, caddr_t), MLEN); 1338 n->m_len = min(MLEN, len + off); 1339 m->m_next = n; 1340 } 1341 m = m->m_next; 1342 } 1343 while (len > 0) { 1344 if (m->m_next == NULL && (len > m->m_len - off)) { 1345 m->m_len += min(len - (m->m_len - off), 1346 M_TRAILINGSPACE(m)); 1347 } 1348 mlen = min (m->m_len - off, len); 1349 bcopy(cp, off + mtod(m, caddr_t), (u_int)mlen); 1350 cp += mlen; 1351 len -= mlen; 1352 mlen += off; 1353 off = 0; 1354 totlen += mlen; 1355 if (len == 0) 1356 break; 1357 if (m->m_next == NULL) { 1358 n = m_get(M_NOWAIT, m->m_type); 1359 if (n == NULL) 1360 break; 1361 n->m_len = min(MLEN, len); 1362 m->m_next = n; 1363 } 1364 m = m->m_next; 1365 } 1366out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen)) 1367 m->m_pkthdr.len = totlen; 1368} 1369 1370/* 1371 * Append the specified data to the indicated mbuf chain, 1372 * Extend the mbuf chain if the new data does not fit in 1373 * existing space. 1374 * 1375 * Return 1 if able to complete the job; otherwise 0. 1376 */ 1377int 1378m_append(struct mbuf *m0, int len, c_caddr_t cp) 1379{ 1380 struct mbuf *m, *n; 1381 int remainder, space; 1382 1383 for (m = m0; m->m_next != NULL; m = m->m_next) 1384 ; 1385 remainder = len; 1386 space = M_TRAILINGSPACE(m); 1387 if (space > 0) { 1388 /* 1389 * Copy into available space. 1390 */ 1391 if (space > remainder) 1392 space = remainder; 1393 bcopy(cp, mtod(m, caddr_t) + m->m_len, space); 1394 m->m_len += space; 1395 cp += space, remainder -= space; 1396 } 1397 while (remainder > 0) { 1398 /* 1399 * Allocate a new mbuf; could check space 1400 * and allocate a cluster instead. 1401 */ 1402 n = m_get(M_NOWAIT, m->m_type); 1403 if (n == NULL) 1404 break; 1405 n->m_len = min(MLEN, remainder); 1406 bcopy(cp, mtod(n, caddr_t), n->m_len); 1407 cp += n->m_len, remainder -= n->m_len; 1408 m->m_next = n; 1409 m = n; 1410 } 1411 if (m0->m_flags & M_PKTHDR) 1412 m0->m_pkthdr.len += len - remainder; 1413 return (remainder == 0); 1414} 1415 1416/* 1417 * Apply function f to the data in an mbuf chain starting "off" bytes from 1418 * the beginning, continuing for "len" bytes. 1419 */ 1420int 1421m_apply(struct mbuf *m, int off, int len, 1422 int (*f)(void *, void *, u_int), void *arg) 1423{ 1424 u_int count; 1425 int rval; 1426 1427 KASSERT(off >= 0, ("m_apply, negative off %d", off)); 1428 KASSERT(len >= 0, ("m_apply, negative len %d", len)); 1429 while (off > 0) { 1430 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain")); 1431 if (off < m->m_len) 1432 break; 1433 off -= m->m_len; 1434 m = m->m_next; 1435 } 1436 while (len > 0) { 1437 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain")); 1438 count = min(m->m_len - off, len); 1439 rval = (*f)(arg, mtod(m, caddr_t) + off, count); 1440 if (rval) 1441 return (rval); 1442 len -= count; 1443 off = 0; 1444 m = m->m_next; 1445 } 1446 return (0); 1447} 1448 1449/* 1450 * Return a pointer to mbuf/offset of location in mbuf chain. 1451 */ 1452struct mbuf * 1453m_getptr(struct mbuf *m, int loc, int *off) 1454{ 1455 1456 while (loc >= 0) { 1457 /* Normal end of search. */ 1458 if (m->m_len > loc) { 1459 *off = loc; 1460 return (m); 1461 } else { 1462 loc -= m->m_len; 1463 if (m->m_next == NULL) { 1464 if (loc == 0) { 1465 /* Point at the end of valid data. */ 1466 *off = m->m_len; 1467 return (m); 1468 } 1469 return (NULL); 1470 } 1471 m = m->m_next; 1472 } 1473 } 1474 return (NULL); 1475} 1476 1477void 1478m_print(const struct mbuf *m, int maxlen) 1479{ 1480 int len; 1481 int pdata; 1482 const struct mbuf *m2; 1483 1484 if (m == NULL) { 1485 printf("mbuf: %p\n", m); 1486 return; 1487 } 1488 1489 if (m->m_flags & M_PKTHDR) 1490 len = m->m_pkthdr.len; 1491 else 1492 len = -1; 1493 m2 = m; 1494 while (m2 != NULL && (len == -1 || len)) { 1495 pdata = m2->m_len; 1496 if (maxlen != -1 && pdata > maxlen) 1497 pdata = maxlen; 1498 printf("mbuf: %p len: %d, next: %p, %b%s", m2, m2->m_len, 1499 m2->m_next, m2->m_flags, "\20\20freelist\17skipfw" 1500 "\11proto5\10proto4\7proto3\6proto2\5proto1\4rdonly" 1501 "\3eor\2pkthdr\1ext", pdata ? "" : "\n"); 1502 if (pdata) 1503 printf(", %*D\n", pdata, (u_char *)m2->m_data, "-"); 1504 if (len != -1) 1505 len -= m2->m_len; 1506 m2 = m2->m_next; 1507 } 1508 if (len > 0) 1509 printf("%d bytes unaccounted for.\n", len); 1510 return; 1511} 1512 1513u_int 1514m_fixhdr(struct mbuf *m0) 1515{ 1516 u_int len; 1517 1518 len = m_length(m0, NULL); 1519 m0->m_pkthdr.len = len; 1520 return (len); 1521} 1522 1523u_int 1524m_length(struct mbuf *m0, struct mbuf **last) 1525{ 1526 struct mbuf *m; 1527 u_int len; 1528 1529 len = 0; 1530 for (m = m0; m != NULL; m = m->m_next) { 1531 len += m->m_len; 1532 if (m->m_next == NULL) 1533 break; 1534 } 1535 if (last != NULL) 1536 *last = m; 1537 return (len); 1538} 1539 1540/* 1541 * Defragment a mbuf chain, returning the shortest possible 1542 * chain of mbufs and clusters. If allocation fails and 1543 * this cannot be completed, NULL will be returned, but 1544 * the passed in chain will be unchanged. Upon success, 1545 * the original chain will be freed, and the new chain 1546 * will be returned. 1547 * 1548 * If a non-packet header is passed in, the original 1549 * mbuf (chain?) will be returned unharmed. 1550 */ 1551struct mbuf * 1552m_defrag(struct mbuf *m0, int how) 1553{ 1554 struct mbuf *m_new = NULL, *m_final = NULL; 1555 int progress = 0, length; 1556 1557 MBUF_CHECKSLEEP(how); 1558 if (!(m0->m_flags & M_PKTHDR)) 1559 return (m0); 1560 1561 m_fixhdr(m0); /* Needed sanity check */ 1562 1563#ifdef MBUF_STRESS_TEST 1564 if (m_defragrandomfailures) { 1565 int temp = arc4random() & 0xff; 1566 if (temp == 0xba) 1567 goto nospace; 1568 } 1569#endif 1570 1571 if (m0->m_pkthdr.len > MHLEN) 1572 m_final = m_getcl(how, MT_DATA, M_PKTHDR); 1573 else 1574 m_final = m_gethdr(how, MT_DATA); 1575 1576 if (m_final == NULL) 1577 goto nospace; 1578 1579 if (m_dup_pkthdr(m_final, m0, how) == 0) 1580 goto nospace; 1581 1582 m_new = m_final; 1583 1584 while (progress < m0->m_pkthdr.len) { 1585 length = m0->m_pkthdr.len - progress; 1586 if (length > MCLBYTES) 1587 length = MCLBYTES; 1588 1589 if (m_new == NULL) { 1590 if (length > MLEN) 1591 m_new = m_getcl(how, MT_DATA, 0); 1592 else 1593 m_new = m_get(how, MT_DATA); 1594 if (m_new == NULL) 1595 goto nospace; 1596 } 1597 1598 m_copydata(m0, progress, length, mtod(m_new, caddr_t)); 1599 progress += length; 1600 m_new->m_len = length; 1601 if (m_new != m_final) 1602 m_cat(m_final, m_new); 1603 m_new = NULL; 1604 } 1605#ifdef MBUF_STRESS_TEST 1606 if (m0->m_next == NULL) 1607 m_defraguseless++; 1608#endif 1609 m_freem(m0); 1610 m0 = m_final; 1611#ifdef MBUF_STRESS_TEST 1612 m_defragpackets++; 1613 m_defragbytes += m0->m_pkthdr.len; 1614#endif 1615 return (m0); 1616nospace: 1617#ifdef MBUF_STRESS_TEST 1618 m_defragfailure++; 1619#endif 1620 if (m_final) 1621 m_freem(m_final); 1622 return (NULL); 1623} 1624 1625/* 1626 * Defragment an mbuf chain, returning at most maxfrags separate 1627 * mbufs+clusters. If this is not possible NULL is returned and 1628 * the original mbuf chain is left in it's present (potentially 1629 * modified) state. We use two techniques: collapsing consecutive 1630 * mbufs and replacing consecutive mbufs by a cluster. 1631 * 1632 * NB: this should really be named m_defrag but that name is taken 1633 */ 1634struct mbuf * 1635m_collapse(struct mbuf *m0, int how, int maxfrags) 1636{ 1637 struct mbuf *m, *n, *n2, **prev; 1638 u_int curfrags; 1639 1640 /* 1641 * Calculate the current number of frags. 1642 */ 1643 curfrags = 0; 1644 for (m = m0; m != NULL; m = m->m_next) 1645 curfrags++; 1646 /* 1647 * First, try to collapse mbufs. Note that we always collapse 1648 * towards the front so we don't need to deal with moving the 1649 * pkthdr. This may be suboptimal if the first mbuf has much 1650 * less data than the following. 1651 */ 1652 m = m0; 1653again: 1654 for (;;) { 1655 n = m->m_next; 1656 if (n == NULL) 1657 break; 1658 if (M_WRITABLE(m) && 1659 n->m_len < M_TRAILINGSPACE(m)) { 1660 bcopy(mtod(n, void *), mtod(m, char *) + m->m_len, 1661 n->m_len); 1662 m->m_len += n->m_len; 1663 m->m_next = n->m_next; 1664 m_free(n); 1665 if (--curfrags <= maxfrags) 1666 return m0; 1667 } else 1668 m = n; 1669 } 1670 KASSERT(maxfrags > 1, 1671 ("maxfrags %u, but normal collapse failed", maxfrags)); 1672 /* 1673 * Collapse consecutive mbufs to a cluster. 1674 */ 1675 prev = &m0->m_next; /* NB: not the first mbuf */ 1676 while ((n = *prev) != NULL) { 1677 if ((n2 = n->m_next) != NULL && 1678 n->m_len + n2->m_len < MCLBYTES) { 1679 m = m_getcl(how, MT_DATA, 0); 1680 if (m == NULL) 1681 goto bad; 1682 bcopy(mtod(n, void *), mtod(m, void *), n->m_len); 1683 bcopy(mtod(n2, void *), mtod(m, char *) + n->m_len, 1684 n2->m_len); 1685 m->m_len = n->m_len + n2->m_len; 1686 m->m_next = n2->m_next; 1687 *prev = m; 1688 m_free(n); 1689 m_free(n2); 1690 if (--curfrags <= maxfrags) /* +1 cl -2 mbufs */ 1691 return m0; 1692 /* 1693 * Still not there, try the normal collapse 1694 * again before we allocate another cluster. 1695 */ 1696 goto again; 1697 } 1698 prev = &n->m_next; 1699 } 1700 /* 1701 * No place where we can collapse to a cluster; punt. 1702 * This can occur if, for example, you request 2 frags 1703 * but the packet requires that both be clusters (we 1704 * never reallocate the first mbuf to avoid moving the 1705 * packet header). 1706 */ 1707bad: 1708 return NULL; 1709} 1710 1711#ifdef MBUF_STRESS_TEST 1712 1713/* 1714 * Fragment an mbuf chain. There's no reason you'd ever want to do 1715 * this in normal usage, but it's great for stress testing various 1716 * mbuf consumers. 1717 * 1718 * If fragmentation is not possible, the original chain will be 1719 * returned. 1720 * 1721 * Possible length values: 1722 * 0 no fragmentation will occur 1723 * > 0 each fragment will be of the specified length 1724 * -1 each fragment will be the same random value in length 1725 * -2 each fragment's length will be entirely random 1726 * (Random values range from 1 to 256) 1727 */ 1728struct mbuf * 1729m_fragment(struct mbuf *m0, int how, int length) 1730{ 1731 struct mbuf *m_new = NULL, *m_final = NULL; 1732 int progress = 0; 1733 1734 if (!(m0->m_flags & M_PKTHDR)) 1735 return (m0); 1736 1737 if ((length == 0) || (length < -2)) 1738 return (m0); 1739 1740 m_fixhdr(m0); /* Needed sanity check */ 1741 1742 m_final = m_getcl(how, MT_DATA, M_PKTHDR); 1743 1744 if (m_final == NULL) 1745 goto nospace; 1746 1747 if (m_dup_pkthdr(m_final, m0, how) == 0) 1748 goto nospace; 1749 1750 m_new = m_final; 1751 1752 if (length == -1) 1753 length = 1 + (arc4random() & 255); 1754 1755 while (progress < m0->m_pkthdr.len) { 1756 int fraglen; 1757 1758 if (length > 0) 1759 fraglen = length; 1760 else 1761 fraglen = 1 + (arc4random() & 255); 1762 if (fraglen > m0->m_pkthdr.len - progress) 1763 fraglen = m0->m_pkthdr.len - progress; 1764 1765 if (fraglen > MCLBYTES) 1766 fraglen = MCLBYTES; 1767 1768 if (m_new == NULL) { 1769 m_new = m_getcl(how, MT_DATA, 0); 1770 if (m_new == NULL) 1771 goto nospace; 1772 } 1773 1774 m_copydata(m0, progress, fraglen, mtod(m_new, caddr_t)); 1775 progress += fraglen; 1776 m_new->m_len = fraglen; 1777 if (m_new != m_final) 1778 m_cat(m_final, m_new); 1779 m_new = NULL; 1780 } 1781 m_freem(m0); 1782 m0 = m_final; 1783 return (m0); 1784nospace: 1785 if (m_final) 1786 m_freem(m_final); 1787 /* Return the original chain on failure */ 1788 return (m0); 1789} 1790 1791#endif 1792 1793/* 1794 * Copy the contents of uio into a properly sized mbuf chain. 1795 */ 1796struct mbuf * 1797m_uiotombuf(struct uio *uio, int how, int len, int align, int flags) 1798{ 1799 struct mbuf *m, *mb; 1800 int error, length; 1801 ssize_t total; 1802 int progress = 0; 1803 1804 /* 1805 * len can be zero or an arbitrary large value bound by 1806 * the total data supplied by the uio. 1807 */ 1808 if (len > 0) 1809 total = min(uio->uio_resid, len); 1810 else 1811 total = uio->uio_resid; 1812 1813 /* 1814 * The smallest unit returned by m_getm2() is a single mbuf 1815 * with pkthdr. We can't align past it. 1816 */ 1817 if (align >= MHLEN) 1818 return (NULL); 1819 1820 /* 1821 * Give us the full allocation or nothing. 1822 * If len is zero return the smallest empty mbuf. 1823 */ 1824 m = m_getm2(NULL, max(total + align, 1), how, MT_DATA, flags); 1825 if (m == NULL) 1826 return (NULL); 1827 m->m_data += align; 1828 1829 /* Fill all mbufs with uio data and update header information. */ 1830 for (mb = m; mb != NULL; mb = mb->m_next) { 1831 length = min(M_TRAILINGSPACE(mb), total - progress); 1832 1833 error = uiomove(mtod(mb, void *), length, uio); 1834 if (error) { 1835 m_freem(m); 1836 return (NULL); 1837 } 1838 1839 mb->m_len = length; 1840 progress += length; 1841 if (flags & M_PKTHDR) 1842 m->m_pkthdr.len += length; 1843 } 1844 KASSERT(progress == total, ("%s: progress != total", __func__)); 1845 1846 return (m); 1847} 1848 1849/* 1850 * Copy an mbuf chain into a uio limited by len if set. 1851 */ 1852int 1853m_mbuftouio(struct uio *uio, struct mbuf *m, int len) 1854{ 1855 int error, length, total; 1856 int progress = 0; 1857 1858 if (len > 0) 1859 total = min(uio->uio_resid, len); 1860 else 1861 total = uio->uio_resid; 1862 1863 /* Fill the uio with data from the mbufs. */ 1864 for (; m != NULL; m = m->m_next) { 1865 length = min(m->m_len, total - progress); 1866 1867 error = uiomove(mtod(m, void *), length, uio); 1868 if (error) 1869 return (error); 1870 1871 progress += length; 1872 } 1873 1874 return (0); 1875} 1876 1877/* 1878 * Set the m_data pointer of a newly-allocated mbuf 1879 * to place an object of the specified size at the 1880 * end of the mbuf, longword aligned. 1881 */ 1882void 1883m_align(struct mbuf *m, int len) 1884{ 1885#ifdef INVARIANTS 1886 const char *msg = "%s: not a virgin mbuf"; 1887#endif 1888 int adjust; 1889 1890 if (m->m_flags & M_EXT) { 1891 KASSERT(m->m_data == m->m_ext.ext_buf, (msg, __func__)); 1892 adjust = m->m_ext.ext_size - len; 1893 } else if (m->m_flags & M_PKTHDR) { 1894 KASSERT(m->m_data == m->m_pktdat, (msg, __func__)); 1895 adjust = MHLEN - len; 1896 } else { 1897 KASSERT(m->m_data == m->m_dat, (msg, __func__)); 1898 adjust = MLEN - len; 1899 } 1900 1901 m->m_data += adjust &~ (sizeof(long)-1); 1902} 1903 1904/* 1905 * Create a writable copy of the mbuf chain. While doing this 1906 * we compact the chain with a goal of producing a chain with 1907 * at most two mbufs. The second mbuf in this chain is likely 1908 * to be a cluster. The primary purpose of this work is to create 1909 * a writable packet for encryption, compression, etc. The 1910 * secondary goal is to linearize the data so the data can be 1911 * passed to crypto hardware in the most efficient manner possible. 1912 */ 1913struct mbuf * 1914m_unshare(struct mbuf *m0, int how) 1915{ 1916 struct mbuf *m, *mprev; 1917 struct mbuf *n, *mfirst, *mlast; 1918 int len, off; 1919 1920 mprev = NULL; 1921 for (m = m0; m != NULL; m = mprev->m_next) { 1922 /* 1923 * Regular mbufs are ignored unless there's a cluster 1924 * in front of it that we can use to coalesce. We do 1925 * the latter mainly so later clusters can be coalesced 1926 * also w/o having to handle them specially (i.e. convert 1927 * mbuf+cluster -> cluster). This optimization is heavily 1928 * influenced by the assumption that we're running over 1929 * Ethernet where MCLBYTES is large enough that the max 1930 * packet size will permit lots of coalescing into a 1931 * single cluster. This in turn permits efficient 1932 * crypto operations, especially when using hardware. 1933 */ 1934 if ((m->m_flags & M_EXT) == 0) { 1935 if (mprev && (mprev->m_flags & M_EXT) && 1936 m->m_len <= M_TRAILINGSPACE(mprev)) { 1937 /* XXX: this ignores mbuf types */ 1938 memcpy(mtod(mprev, caddr_t) + mprev->m_len, 1939 mtod(m, caddr_t), m->m_len); 1940 mprev->m_len += m->m_len; 1941 mprev->m_next = m->m_next; /* unlink from chain */ 1942 m_free(m); /* reclaim mbuf */ 1943#if 0 1944 newipsecstat.ips_mbcoalesced++; 1945#endif 1946 } else { 1947 mprev = m; 1948 } 1949 continue; 1950 } 1951 /* 1952 * Writable mbufs are left alone (for now). 1953 */ 1954 if (M_WRITABLE(m)) { 1955 mprev = m; 1956 continue; 1957 } 1958 1959 /* 1960 * Not writable, replace with a copy or coalesce with 1961 * the previous mbuf if possible (since we have to copy 1962 * it anyway, we try to reduce the number of mbufs and 1963 * clusters so that future work is easier). 1964 */ 1965 KASSERT(m->m_flags & M_EXT, ("m_flags 0x%x", m->m_flags)); 1966 /* NB: we only coalesce into a cluster or larger */ 1967 if (mprev != NULL && (mprev->m_flags & M_EXT) && 1968 m->m_len <= M_TRAILINGSPACE(mprev)) { 1969 /* XXX: this ignores mbuf types */ 1970 memcpy(mtod(mprev, caddr_t) + mprev->m_len, 1971 mtod(m, caddr_t), m->m_len); 1972 mprev->m_len += m->m_len; 1973 mprev->m_next = m->m_next; /* unlink from chain */ 1974 m_free(m); /* reclaim mbuf */ 1975#if 0 1976 newipsecstat.ips_clcoalesced++; 1977#endif 1978 continue; 1979 } 1980 1981 /* 1982 * Allocate new space to hold the copy and copy the data. 1983 * We deal with jumbo mbufs (i.e. m_len > MCLBYTES) by 1984 * splitting them into clusters. We could just malloc a 1985 * buffer and make it external but too many device drivers 1986 * don't know how to break up the non-contiguous memory when 1987 * doing DMA. 1988 */ 1989 n = m_getcl(how, m->m_type, m->m_flags); 1990 if (n == NULL) { 1991 m_freem(m0); 1992 return (NULL); 1993 } 1994 len = m->m_len; 1995 off = 0; 1996 mfirst = n; 1997 mlast = NULL; 1998 for (;;) { 1999 int cc = min(len, MCLBYTES); 2000 memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + off, cc); 2001 n->m_len = cc; 2002 if (mlast != NULL) 2003 mlast->m_next = n; 2004 mlast = n; 2005#if 0 2006 newipsecstat.ips_clcopied++; 2007#endif 2008 2009 len -= cc; 2010 if (len <= 0) 2011 break; 2012 off += cc; 2013 2014 n = m_getcl(how, m->m_type, m->m_flags); 2015 if (n == NULL) { 2016 m_freem(mfirst); 2017 m_freem(m0); 2018 return (NULL); 2019 } 2020 } 2021 n->m_next = m->m_next; 2022 if (mprev == NULL) 2023 m0 = mfirst; /* new head of chain */ 2024 else 2025 mprev->m_next = mfirst; /* replace old mbuf */ 2026 m_free(m); /* release old mbuf */ 2027 mprev = mfirst; 2028 } 2029 return (m0); 2030} 2031 2032#ifdef MBUF_PROFILING 2033 2034#define MP_BUCKETS 32 /* don't just change this as things may overflow.*/ 2035struct mbufprofile { 2036 uintmax_t wasted[MP_BUCKETS]; 2037 uintmax_t used[MP_BUCKETS]; 2038 uintmax_t segments[MP_BUCKETS]; 2039} mbprof; 2040 2041#define MP_MAXDIGITS 21 /* strlen("16,000,000,000,000,000,000") == 21 */ 2042#define MP_NUMLINES 6 2043#define MP_NUMSPERLINE 16 2044#define MP_EXTRABYTES 64 /* > strlen("used:\nwasted:\nsegments:\n") */ 2045/* work out max space needed and add a bit of spare space too */ 2046#define MP_MAXLINE ((MP_MAXDIGITS+1) * MP_NUMSPERLINE) 2047#define MP_BUFSIZE ((MP_MAXLINE * MP_NUMLINES) + 1 + MP_EXTRABYTES) 2048 2049char mbprofbuf[MP_BUFSIZE]; 2050 2051void 2052m_profile(struct mbuf *m) 2053{ 2054 int segments = 0; 2055 int used = 0; 2056 int wasted = 0; 2057 2058 while (m) { 2059 segments++; 2060 used += m->m_len; 2061 if (m->m_flags & M_EXT) { 2062 wasted += MHLEN - sizeof(m->m_ext) + 2063 m->m_ext.ext_size - m->m_len; 2064 } else { 2065 if (m->m_flags & M_PKTHDR) 2066 wasted += MHLEN - m->m_len; 2067 else 2068 wasted += MLEN - m->m_len; 2069 } 2070 m = m->m_next; 2071 } 2072 /* be paranoid.. it helps */ 2073 if (segments > MP_BUCKETS - 1) 2074 segments = MP_BUCKETS - 1; 2075 if (used > 100000) 2076 used = 100000; 2077 if (wasted > 100000) 2078 wasted = 100000; 2079 /* store in the appropriate bucket */ 2080 /* don't bother locking. if it's slightly off, so what? */ 2081 mbprof.segments[segments]++; 2082 mbprof.used[fls(used)]++; 2083 mbprof.wasted[fls(wasted)]++; 2084} 2085 2086static void 2087mbprof_textify(void) 2088{ 2089 int offset; 2090 char *c; 2091 uint64_t *p; 2092 2093 2094 p = &mbprof.wasted[0]; 2095 c = mbprofbuf; 2096 offset = snprintf(c, MP_MAXLINE + 10, 2097 "wasted:\n" 2098 "%ju %ju %ju %ju %ju %ju %ju %ju " 2099 "%ju %ju %ju %ju %ju %ju %ju %ju\n", 2100 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7], 2101 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]); 2102#ifdef BIG_ARRAY 2103 p = &mbprof.wasted[16]; 2104 c += offset; 2105 offset = snprintf(c, MP_MAXLINE, 2106 "%ju %ju %ju %ju %ju %ju %ju %ju " 2107 "%ju %ju %ju %ju %ju %ju %ju %ju\n", 2108 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7], 2109 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]); 2110#endif 2111 p = &mbprof.used[0]; 2112 c += offset; 2113 offset = snprintf(c, MP_MAXLINE + 10, 2114 "used:\n" 2115 "%ju %ju %ju %ju %ju %ju %ju %ju " 2116 "%ju %ju %ju %ju %ju %ju %ju %ju\n", 2117 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7], 2118 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]); 2119#ifdef BIG_ARRAY 2120 p = &mbprof.used[16]; 2121 c += offset; 2122 offset = snprintf(c, MP_MAXLINE, 2123 "%ju %ju %ju %ju %ju %ju %ju %ju " 2124 "%ju %ju %ju %ju %ju %ju %ju %ju\n", 2125 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7], 2126 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]); 2127#endif 2128 p = &mbprof.segments[0]; 2129 c += offset; 2130 offset = snprintf(c, MP_MAXLINE + 10, 2131 "segments:\n" 2132 "%ju %ju %ju %ju %ju %ju %ju %ju " 2133 "%ju %ju %ju %ju %ju %ju %ju %ju\n", 2134 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7], 2135 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]); 2136#ifdef BIG_ARRAY 2137 p = &mbprof.segments[16]; 2138 c += offset; 2139 offset = snprintf(c, MP_MAXLINE, 2140 "%ju %ju %ju %ju %ju %ju %ju %ju " 2141 "%ju %ju %ju %ju %ju %ju %ju %jju", 2142 p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7], 2143 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]); 2144#endif 2145} 2146 2147static int 2148mbprof_handler(SYSCTL_HANDLER_ARGS) 2149{ 2150 int error; 2151 2152 mbprof_textify(); 2153 error = SYSCTL_OUT(req, mbprofbuf, strlen(mbprofbuf) + 1); 2154 return (error); 2155} 2156 2157static int 2158mbprof_clr_handler(SYSCTL_HANDLER_ARGS) 2159{ 2160 int clear, error; 2161 2162 clear = 0; 2163 error = sysctl_handle_int(oidp, &clear, 0, req); 2164 if (error || !req->newptr) 2165 return (error); 2166 2167 if (clear) { 2168 bzero(&mbprof, sizeof(mbprof)); 2169 } 2170 2171 return (error); 2172} 2173 2174 2175SYSCTL_PROC(_kern_ipc, OID_AUTO, mbufprofile, CTLTYPE_STRING|CTLFLAG_RD, 2176 NULL, 0, mbprof_handler, "A", "mbuf profiling statistics"); 2177 2178SYSCTL_PROC(_kern_ipc, OID_AUTO, mbufprofileclr, CTLTYPE_INT|CTLFLAG_RW, 2179 NULL, 0, mbprof_clr_handler, "I", "clear mbuf profiling statistics"); 2180#endif 2181 2182