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