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 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * @(#)uipc_mbuf.c 8.2 (Berkeley) 1/4/94 34 * $FreeBSD: src/sys/kern/uipc_mbuf.c,v 1.51.2.24 2003/04/15 06:59:29 silby Exp $ 35 * $DragonFly: src/sys/kern/uipc_mbuf.c,v 1.2 2003/06/17 04:28:41 dillon Exp $ 36 */ 37 38 #include "opt_param.h" 39 #include "opt_mbuf_stress_test.h" 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/malloc.h> 43 #include <sys/mbuf.h> 44 #include <sys/kernel.h> 45 #include <sys/sysctl.h> 46 #include <sys/domain.h> 47 #include <sys/protosw.h> 48 49 #include <vm/vm.h> 50 #include <vm/vm_kern.h> 51 #include <vm/vm_extern.h> 52 53 #ifdef INVARIANTS 54 #include <machine/cpu.h> 55 #endif 56 57 static void mbinit __P((void *)); 58 SYSINIT(mbuf, SI_SUB_MBUF, SI_ORDER_FIRST, mbinit, NULL) 59 60 struct mbuf *mbutl; 61 char *mclrefcnt; 62 struct mbstat mbstat; 63 u_long mbtypes[MT_NTYPES]; 64 struct mbuf *mmbfree; 65 union mcluster *mclfree; 66 int max_linkhdr; 67 int max_protohdr; 68 int max_hdr; 69 int max_datalen; 70 int m_defragpackets; 71 int m_defragbytes; 72 int m_defraguseless; 73 int m_defragfailure; 74 #ifdef MBUF_STRESS_TEST 75 int m_defragrandomfailures; 76 #endif 77 78 int nmbclusters; 79 int nmbufs; 80 u_int m_mballoc_wid = 0; 81 u_int m_clalloc_wid = 0; 82 83 SYSCTL_DECL(_kern_ipc); 84 SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RW, 85 &max_linkhdr, 0, ""); 86 SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RW, 87 &max_protohdr, 0, ""); 88 SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RW, &max_hdr, 0, ""); 89 SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RW, 90 &max_datalen, 0, ""); 91 SYSCTL_INT(_kern_ipc, OID_AUTO, mbuf_wait, CTLFLAG_RW, 92 &mbuf_wait, 0, ""); 93 SYSCTL_STRUCT(_kern_ipc, KIPC_MBSTAT, mbstat, CTLFLAG_RW, &mbstat, mbstat, ""); 94 SYSCTL_OPAQUE(_kern_ipc, OID_AUTO, mbtypes, CTLFLAG_RD, mbtypes, 95 sizeof(mbtypes), "LU", ""); 96 SYSCTL_INT(_kern_ipc, KIPC_NMBCLUSTERS, nmbclusters, CTLFLAG_RD, 97 &nmbclusters, 0, "Maximum number of mbuf clusters available"); 98 SYSCTL_INT(_kern_ipc, OID_AUTO, nmbufs, CTLFLAG_RD, &nmbufs, 0, 99 "Maximum number of mbufs available"); 100 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD, 101 &m_defragpackets, 0, ""); 102 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD, 103 &m_defragbytes, 0, ""); 104 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD, 105 &m_defraguseless, 0, ""); 106 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD, 107 &m_defragfailure, 0, ""); 108 #ifdef MBUF_STRESS_TEST 109 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW, 110 &m_defragrandomfailures, 0, ""); 111 #endif 112 113 static void m_reclaim __P((void)); 114 115 #ifndef NMBCLUSTERS 116 #define NMBCLUSTERS (512 + maxusers * 16) 117 #endif 118 #ifndef NMBUFS 119 #define NMBUFS (nmbclusters * 4) 120 #endif 121 122 /* 123 * Perform sanity checks of tunables declared above. 124 */ 125 static void 126 tunable_mbinit(void *dummy) 127 { 128 129 /* 130 * This has to be done before VM init. 131 */ 132 nmbclusters = NMBCLUSTERS; 133 TUNABLE_INT_FETCH("kern.ipc.nmbclusters", &nmbclusters); 134 nmbufs = NMBUFS; 135 TUNABLE_INT_FETCH("kern.ipc.nmbufs", &nmbufs); 136 /* Sanity checks */ 137 if (nmbufs < nmbclusters * 2) 138 nmbufs = nmbclusters * 2; 139 140 return; 141 } 142 SYSINIT(tunable_mbinit, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_mbinit, NULL); 143 144 /* "number of clusters of pages" */ 145 #define NCL_INIT 1 146 147 #define NMB_INIT 16 148 149 /* ARGSUSED*/ 150 static void 151 mbinit(dummy) 152 void *dummy; 153 { 154 int s; 155 156 mmbfree = NULL; mclfree = NULL; 157 mbstat.m_msize = MSIZE; 158 mbstat.m_mclbytes = MCLBYTES; 159 mbstat.m_minclsize = MINCLSIZE; 160 mbstat.m_mlen = MLEN; 161 mbstat.m_mhlen = MHLEN; 162 163 s = splimp(); 164 if (m_mballoc(NMB_INIT, M_DONTWAIT) == 0) 165 goto bad; 166 #if MCLBYTES <= PAGE_SIZE 167 if (m_clalloc(NCL_INIT, M_DONTWAIT) == 0) 168 goto bad; 169 #else 170 /* It's OK to call contigmalloc in this context. */ 171 if (m_clalloc(16, M_WAIT) == 0) 172 goto bad; 173 #endif 174 splx(s); 175 return; 176 bad: 177 panic("mbinit"); 178 } 179 180 /* 181 * Allocate at least nmb mbufs and place on mbuf free list. 182 * Must be called at splimp. 183 */ 184 /* ARGSUSED */ 185 int 186 m_mballoc(nmb, how) 187 register int nmb; 188 int how; 189 { 190 register caddr_t p; 191 register int i; 192 int nbytes; 193 194 /* 195 * If we've hit the mbuf limit, stop allocating from mb_map, 196 * (or trying to) in order to avoid dipping into the section of 197 * mb_map which we've "reserved" for clusters. 198 */ 199 if ((nmb + mbstat.m_mbufs) > nmbufs) 200 return (0); 201 202 /* 203 * Once we run out of map space, it will be impossible to get 204 * any more (nothing is ever freed back to the map) 205 * -- however you are not dead as m_reclaim might 206 * still be able to free a substantial amount of space. 207 * 208 * XXX Furthermore, we can also work with "recycled" mbufs (when 209 * we're calling with M_WAIT the sleep procedure will be woken 210 * up when an mbuf is freed. See m_mballoc_wait()). 211 */ 212 if (mb_map_full) 213 return (0); 214 215 nbytes = round_page(nmb * MSIZE); 216 p = (caddr_t)kmem_malloc(mb_map, nbytes, M_NOWAIT); 217 if (p == 0 && how == M_WAIT) { 218 mbstat.m_wait++; 219 p = (caddr_t)kmem_malloc(mb_map, nbytes, M_WAITOK); 220 } 221 222 /* 223 * Either the map is now full, or `how' is M_NOWAIT and there 224 * are no pages left. 225 */ 226 if (p == NULL) 227 return (0); 228 229 nmb = nbytes / MSIZE; 230 for (i = 0; i < nmb; i++) { 231 ((struct mbuf *)p)->m_next = mmbfree; 232 mmbfree = (struct mbuf *)p; 233 p += MSIZE; 234 } 235 mbstat.m_mbufs += nmb; 236 mbtypes[MT_FREE] += nmb; 237 return (1); 238 } 239 240 /* 241 * Once the mb_map has been exhausted and if the call to the allocation macros 242 * (or, in some cases, functions) is with M_WAIT, then it is necessary to rely 243 * solely on reclaimed mbufs. Here we wait for an mbuf to be freed for a 244 * designated (mbuf_wait) time. 245 */ 246 struct mbuf * 247 m_mballoc_wait(int caller, int type) 248 { 249 struct mbuf *p; 250 int s; 251 252 s = splimp(); 253 m_mballoc_wid++; 254 if ((tsleep(&m_mballoc_wid, PVM, "mballc", mbuf_wait)) == EWOULDBLOCK) 255 m_mballoc_wid--; 256 splx(s); 257 258 /* 259 * Now that we (think) that we've got something, we will redo an 260 * MGET, but avoid getting into another instance of m_mballoc_wait() 261 * XXX: We retry to fetch _even_ if the sleep timed out. This is left 262 * this way, purposely, in the [unlikely] case that an mbuf was 263 * freed but the sleep was not awakened in time. 264 */ 265 p = NULL; 266 switch (caller) { 267 case MGET_C: 268 MGET(p, M_DONTWAIT, type); 269 break; 270 case MGETHDR_C: 271 MGETHDR(p, M_DONTWAIT, type); 272 break; 273 default: 274 panic("m_mballoc_wait: invalid caller (%d)", caller); 275 } 276 277 s = splimp(); 278 if (p != NULL) { /* We waited and got something... */ 279 mbstat.m_wait++; 280 /* Wake up another if we have more free. */ 281 if (mmbfree != NULL) 282 MMBWAKEUP(); 283 } 284 splx(s); 285 return (p); 286 } 287 288 #if MCLBYTES > PAGE_SIZE 289 static int i_want_my_mcl; 290 291 static void 292 kproc_mclalloc(void) 293 { 294 int status; 295 296 while (1) { 297 tsleep(&i_want_my_mcl, PVM, "mclalloc", 0); 298 299 for (; i_want_my_mcl; i_want_my_mcl--) { 300 if (m_clalloc(1, M_WAIT) == 0) 301 printf("m_clalloc failed even in process context!\n"); 302 } 303 } 304 } 305 306 static struct proc *mclallocproc; 307 static struct kproc_desc mclalloc_kp = { 308 "mclalloc", 309 kproc_mclalloc, 310 &mclallocproc 311 }; 312 SYSINIT(mclallocproc, SI_SUB_KTHREAD_UPDATE, SI_ORDER_ANY, kproc_start, 313 &mclalloc_kp); 314 #endif 315 316 /* 317 * Allocate some number of mbuf clusters 318 * and place on cluster free list. 319 * Must be called at splimp. 320 */ 321 /* ARGSUSED */ 322 int 323 m_clalloc(ncl, how) 324 register int ncl; 325 int how; 326 { 327 register caddr_t p; 328 register int i; 329 int npg; 330 331 /* 332 * If we've hit the mcluster number limit, stop allocating from 333 * mb_map, (or trying to) in order to avoid dipping into the section 334 * of mb_map which we've "reserved" for mbufs. 335 */ 336 if ((ncl + mbstat.m_clusters) > nmbclusters) 337 goto m_clalloc_fail; 338 339 /* 340 * Once we run out of map space, it will be impossible 341 * to get any more (nothing is ever freed back to the 342 * map). From this point on, we solely rely on freed 343 * mclusters. 344 */ 345 if (mb_map_full) 346 goto m_clalloc_fail; 347 348 #if MCLBYTES > PAGE_SIZE 349 if (how != M_WAIT) { 350 i_want_my_mcl += ncl; 351 wakeup(&i_want_my_mcl); 352 mbstat.m_wait++; 353 p = 0; 354 } else { 355 p = contigmalloc1(MCLBYTES * ncl, M_DEVBUF, M_WAITOK, 0ul, 356 ~0ul, PAGE_SIZE, 0, mb_map); 357 } 358 #else 359 npg = ncl; 360 p = (caddr_t)kmem_malloc(mb_map, ctob(npg), 361 how != M_WAIT ? M_NOWAIT : M_WAITOK); 362 ncl = ncl * PAGE_SIZE / MCLBYTES; 363 #endif 364 /* 365 * Either the map is now full, or `how' is M_NOWAIT and there 366 * are no pages left. 367 */ 368 if (p == NULL) { 369 static int last_report ; /* when we did that (in ticks) */ 370 m_clalloc_fail: 371 mbstat.m_drops++; 372 if (ticks < last_report || (ticks - last_report) >= hz) { 373 last_report = ticks; 374 printf("All mbuf clusters exhausted, please see tuning(7).\n"); 375 } 376 return (0); 377 } 378 379 for (i = 0; i < ncl; i++) { 380 ((union mcluster *)p)->mcl_next = mclfree; 381 mclfree = (union mcluster *)p; 382 p += MCLBYTES; 383 mbstat.m_clfree++; 384 } 385 mbstat.m_clusters += ncl; 386 return (1); 387 } 388 389 /* 390 * Once the mb_map submap has been exhausted and the allocation is called with 391 * M_WAIT, we rely on the mclfree union pointers. If nothing is free, we will 392 * sleep for a designated amount of time (mbuf_wait) or until we're woken up 393 * due to sudden mcluster availability. 394 */ 395 caddr_t 396 m_clalloc_wait(void) 397 { 398 caddr_t p; 399 int s; 400 401 #ifdef __i386__ 402 /* If in interrupt context, and INVARIANTS, maintain sanity and die. */ 403 KASSERT(intr_nesting_level == 0, ("CLALLOC: CANNOT WAIT IN INTERRUPT")); 404 #endif 405 406 /* Sleep until something's available or until we expire. */ 407 m_clalloc_wid++; 408 if ((tsleep(&m_clalloc_wid, PVM, "mclalc", mbuf_wait)) == EWOULDBLOCK) 409 m_clalloc_wid--; 410 411 /* 412 * Now that we (think) that we've got something, we will redo and 413 * MGET, but avoid getting into another instance of m_clalloc_wait() 414 */ 415 p = NULL; 416 MCLALLOC(p, M_DONTWAIT); 417 418 s = splimp(); 419 if (p != NULL) { /* We waited and got something... */ 420 mbstat.m_wait++; 421 /* Wake up another if we have more free. */ 422 if (mclfree != NULL) 423 MCLWAKEUP(); 424 } 425 426 splx(s); 427 return (p); 428 } 429 430 /* 431 * When MGET fails, ask protocols to free space when short of memory, 432 * then re-attempt to allocate an mbuf. 433 */ 434 struct mbuf * 435 m_retry(i, t) 436 int i, t; 437 { 438 register struct mbuf *m; 439 440 /* 441 * Must only do the reclaim if not in an interrupt context. 442 */ 443 if (i == M_WAIT) { 444 #ifdef __i386__ 445 KASSERT(intr_nesting_level == 0, 446 ("MBALLOC: CANNOT WAIT IN INTERRUPT")); 447 #endif 448 m_reclaim(); 449 } 450 451 /* 452 * Both m_mballoc_wait and m_retry must be nulled because 453 * when the MGET macro is run from here, we deffinately do _not_ 454 * want to enter an instance of m_mballoc_wait() or m_retry() (again!) 455 */ 456 #define m_mballoc_wait(caller,type) (struct mbuf *)0 457 #define m_retry(i, t) (struct mbuf *)0 458 MGET(m, i, t); 459 #undef m_retry 460 #undef m_mballoc_wait 461 462 if (m != NULL) 463 mbstat.m_wait++; 464 else { 465 static int last_report ; /* when we did that (in ticks) */ 466 mbstat.m_drops++; 467 if (ticks < last_report || (ticks - last_report) >= hz) { 468 last_report = ticks; 469 printf("All mbufs exhausted, please see tuning(7).\n"); 470 } 471 } 472 473 return (m); 474 } 475 476 /* 477 * As above; retry an MGETHDR. 478 */ 479 struct mbuf * 480 m_retryhdr(i, t) 481 int i, t; 482 { 483 register struct mbuf *m; 484 485 /* 486 * Must only do the reclaim if not in an interrupt context. 487 */ 488 if (i == M_WAIT) { 489 #ifdef __i386__ 490 KASSERT(intr_nesting_level == 0, 491 ("MBALLOC: CANNOT WAIT IN INTERRUPT")); 492 #endif 493 m_reclaim(); 494 } 495 496 #define m_mballoc_wait(caller,type) (struct mbuf *)0 497 #define m_retryhdr(i, t) (struct mbuf *)0 498 MGETHDR(m, i, t); 499 #undef m_retryhdr 500 #undef m_mballoc_wait 501 502 if (m != NULL) 503 mbstat.m_wait++; 504 else { 505 static int last_report ; /* when we did that (in ticks) */ 506 mbstat.m_drops++; 507 if (ticks < last_report || (ticks - last_report) >= hz) { 508 last_report = ticks; 509 printf("All mbufs exhausted, please see tuning(7).\n"); 510 } 511 } 512 513 return (m); 514 } 515 516 static void 517 m_reclaim() 518 { 519 register struct domain *dp; 520 register struct protosw *pr; 521 int s = splimp(); 522 523 for (dp = domains; dp; dp = dp->dom_next) 524 for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) 525 if (pr->pr_drain) 526 (*pr->pr_drain)(); 527 splx(s); 528 mbstat.m_drain++; 529 } 530 531 /* 532 * Space allocation routines. 533 * These are also available as macros 534 * for critical paths. 535 */ 536 struct mbuf * 537 m_get(how, type) 538 int how, type; 539 { 540 register struct mbuf *m; 541 542 MGET(m, how, type); 543 return (m); 544 } 545 546 struct mbuf * 547 m_gethdr(how, type) 548 int how, type; 549 { 550 register struct mbuf *m; 551 552 MGETHDR(m, how, type); 553 return (m); 554 } 555 556 struct mbuf * 557 m_getclr(how, type) 558 int how, type; 559 { 560 register struct mbuf *m; 561 562 MGET(m, how, type); 563 if (m == 0) 564 return (0); 565 bzero(mtod(m, caddr_t), MLEN); 566 return (m); 567 } 568 569 /* 570 * m_getcl() returns an mbuf with an attached cluster. 571 * Because many network drivers use this kind of buffers a lot, it is 572 * convenient to keep a small pool of free buffers of this kind. 573 * Even a small size such as 10 gives about 10% improvement in the 574 * forwarding rate in a bridge or router. 575 * The size of this free list is controlled by the sysctl variable 576 * mcl_pool_max. The list is populated on m_freem(), and used in 577 * m_getcl() if elements are available. 578 */ 579 static struct mbuf *mcl_pool; 580 static int mcl_pool_now; 581 static int mcl_pool_max = 0; 582 583 SYSCTL_INT(_kern_ipc, OID_AUTO, mcl_pool_max, CTLFLAG_RW, &mcl_pool_max, 0, 584 "Maximum number of mbufs+cluster in free list"); 585 SYSCTL_INT(_kern_ipc, OID_AUTO, mcl_pool_now, CTLFLAG_RD, &mcl_pool_now, 0, 586 "Current number of mbufs+cluster in free list"); 587 588 struct mbuf * 589 m_getcl(int how, short type, int flags) 590 { 591 int s = splimp(); 592 struct mbuf *mp; 593 594 if (flags & M_PKTHDR) { 595 if (type == MT_DATA && mcl_pool) { 596 mp = mcl_pool; 597 mcl_pool = mp->m_nextpkt; 598 mcl_pool_now--; 599 splx(s); 600 mp->m_nextpkt = NULL; 601 mp->m_data = mp->m_ext.ext_buf; 602 mp->m_flags = M_PKTHDR|M_EXT; 603 mp->m_pkthdr.rcvif = NULL; 604 mp->m_pkthdr.csum_flags = 0; 605 return mp; 606 } else 607 MGETHDR(mp, how, type); 608 } else 609 MGET(mp, how, type); 610 if (mp) { 611 MCLGET(mp, how); 612 if ( (mp->m_flags & M_EXT) == 0) { 613 m_free(mp); 614 mp = NULL; 615 } 616 } 617 splx(s); 618 return mp; 619 } 620 621 /* 622 * struct mbuf * 623 * m_getm(m, len, how, type) 624 * 625 * This will allocate len-worth of mbufs and/or mbuf clusters (whatever fits 626 * best) and return a pointer to the top of the allocated chain. If m is 627 * non-null, then we assume that it is a single mbuf or an mbuf chain to 628 * which we want len bytes worth of mbufs and/or clusters attached, and so 629 * if we succeed in allocating it, we will just return a pointer to m. 630 * 631 * If we happen to fail at any point during the allocation, we will free 632 * up everything we have already allocated and return NULL. 633 * 634 */ 635 struct mbuf * 636 m_getm(struct mbuf *m, int len, int how, int type) 637 { 638 struct mbuf *top, *tail, *mp, *mtail = NULL; 639 640 KASSERT(len >= 0, ("len is < 0 in m_getm")); 641 642 MGET(mp, how, type); 643 if (mp == NULL) 644 return (NULL); 645 else if (len > MINCLSIZE) { 646 MCLGET(mp, how); 647 if ((mp->m_flags & M_EXT) == 0) { 648 m_free(mp); 649 return (NULL); 650 } 651 } 652 mp->m_len = 0; 653 len -= M_TRAILINGSPACE(mp); 654 655 if (m != NULL) 656 for (mtail = m; mtail->m_next != NULL; mtail = mtail->m_next); 657 else 658 m = mp; 659 660 top = tail = mp; 661 while (len > 0) { 662 MGET(mp, how, type); 663 if (mp == NULL) 664 goto failed; 665 666 tail->m_next = mp; 667 tail = mp; 668 if (len > MINCLSIZE) { 669 MCLGET(mp, how); 670 if ((mp->m_flags & M_EXT) == 0) 671 goto failed; 672 } 673 674 mp->m_len = 0; 675 len -= M_TRAILINGSPACE(mp); 676 } 677 678 if (mtail != NULL) 679 mtail->m_next = top; 680 return (m); 681 682 failed: 683 m_freem(top); 684 return (NULL); 685 } 686 687 /* 688 * MFREE(struct mbuf *m, struct mbuf *n) 689 * Free a single mbuf and associated external storage. 690 * Place the successor, if any, in n. 691 * 692 * we do need to check non-first mbuf for m_aux, since some of existing 693 * code does not call M_PREPEND properly. 694 * (example: call to bpf_mtap from drivers) 695 */ 696 #define MFREE(m, n) MBUFLOCK( \ 697 struct mbuf *_mm = (m); \ 698 \ 699 KASSERT(_mm->m_type != MT_FREE, ("freeing free mbuf")); \ 700 mbtypes[_mm->m_type]--; \ 701 if ((_mm->m_flags & M_PKTHDR) != 0) \ 702 m_tag_delete_chain(_mm, NULL); \ 703 if (_mm->m_flags & M_EXT) \ 704 MEXTFREE1(m); \ 705 (n) = _mm->m_next; \ 706 _mm->m_type = MT_FREE; \ 707 mbtypes[MT_FREE]++; \ 708 _mm->m_next = mmbfree; \ 709 mmbfree = _mm; \ 710 MMBWAKEUP(); \ 711 ) 712 713 struct mbuf * 714 m_free(m) 715 struct mbuf *m; 716 { 717 register struct mbuf *n; 718 719 MFREE(m, n); 720 return (n); 721 } 722 723 void 724 m_freem(m) 725 struct mbuf *m; 726 { 727 int s = splimp(); 728 729 /* 730 * Try to keep a small pool of mbuf+cluster for quick use in 731 * device drivers. A good candidate is a M_PKTHDR buffer with 732 * only one cluster attached. Other mbufs, or those exceeding 733 * the pool size, are just m_free'd in the usual way. 734 * The following code makes sure that m_next, m_type, 735 * m_pkthdr.aux and m_ext.* are properly initialized. 736 * Other fields in the mbuf are initialized in m_getcl() 737 * upon allocation. 738 */ 739 if (mcl_pool_now < mcl_pool_max && m && m->m_next == NULL && 740 (m->m_flags & (M_PKTHDR|M_EXT)) == (M_PKTHDR|M_EXT) && 741 m->m_type == MT_DATA && M_EXT_WRITABLE(m) ) { 742 m_tag_delete_chain(m, NULL); 743 m->m_nextpkt = mcl_pool; 744 mcl_pool = m; 745 mcl_pool_now++; 746 } else { 747 while (m) 748 m = m_free(m); 749 } 750 splx(s); 751 } 752 753 /* 754 * Mbuffer utility routines. 755 */ 756 757 /* 758 * Lesser-used path for M_PREPEND: 759 * allocate new mbuf to prepend to chain, 760 * copy junk along. 761 */ 762 struct mbuf * 763 m_prepend(m, len, how) 764 register struct mbuf *m; 765 int len, how; 766 { 767 struct mbuf *mn; 768 769 MGET(mn, how, m->m_type); 770 if (mn == (struct mbuf *)NULL) { 771 m_freem(m); 772 return ((struct mbuf *)NULL); 773 } 774 if (m->m_flags & M_PKTHDR) 775 M_MOVE_PKTHDR(mn, m); 776 mn->m_next = m; 777 m = mn; 778 if (len < MHLEN) 779 MH_ALIGN(m, len); 780 m->m_len = len; 781 return (m); 782 } 783 784 /* 785 * Make a copy of an mbuf chain starting "off0" bytes from the beginning, 786 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf. 787 * The wait parameter is a choice of M_WAIT/M_DONTWAIT from caller. 788 * Note that the copy is read-only, because clusters are not copied, 789 * only their reference counts are incremented. 790 */ 791 #define MCFail (mbstat.m_mcfail) 792 793 struct mbuf * 794 m_copym(m, off0, len, wait) 795 register struct mbuf *m; 796 int off0, wait; 797 register int len; 798 { 799 register struct mbuf *n, **np; 800 register int off = off0; 801 struct mbuf *top; 802 int copyhdr = 0; 803 804 KASSERT(off >= 0, ("m_copym, negative off %d", off)); 805 KASSERT(len >= 0, ("m_copym, negative len %d", len)); 806 if (off == 0 && m->m_flags & M_PKTHDR) 807 copyhdr = 1; 808 while (off > 0) { 809 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain")); 810 if (off < m->m_len) 811 break; 812 off -= m->m_len; 813 m = m->m_next; 814 } 815 np = ⊤ 816 top = 0; 817 while (len > 0) { 818 if (m == 0) { 819 KASSERT(len == M_COPYALL, 820 ("m_copym, length > size of mbuf chain")); 821 break; 822 } 823 MGET(n, wait, m->m_type); 824 *np = n; 825 if (n == 0) 826 goto nospace; 827 if (copyhdr) { 828 if (!m_dup_pkthdr(n, m, wait)) 829 goto nospace; 830 if (len == M_COPYALL) 831 n->m_pkthdr.len -= off0; 832 else 833 n->m_pkthdr.len = len; 834 copyhdr = 0; 835 } 836 n->m_len = min(len, m->m_len - off); 837 if (m->m_flags & M_EXT) { 838 n->m_data = m->m_data + off; 839 if (m->m_ext.ext_ref == NULL) { 840 atomic_add_char( 841 &mclrefcnt[mtocl(m->m_ext.ext_buf)], 1); 842 } else { 843 int s = splimp(); 844 845 (*m->m_ext.ext_ref)(m->m_ext.ext_buf, 846 m->m_ext.ext_size); 847 splx(s); 848 } 849 n->m_ext = m->m_ext; 850 n->m_flags |= M_EXT; 851 } else 852 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t), 853 (unsigned)n->m_len); 854 if (len != M_COPYALL) 855 len -= n->m_len; 856 off = 0; 857 m = m->m_next; 858 np = &n->m_next; 859 } 860 if (top == 0) 861 MCFail++; 862 return (top); 863 nospace: 864 m_freem(top); 865 MCFail++; 866 return (0); 867 } 868 869 /* 870 * Copy an entire packet, including header (which must be present). 871 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'. 872 * Note that the copy is read-only, because clusters are not copied, 873 * only their reference counts are incremented. 874 * Preserve alignment of the first mbuf so if the creator has left 875 * some room at the beginning (e.g. for inserting protocol headers) 876 * the copies also have the room available. 877 */ 878 struct mbuf * 879 m_copypacket(m, how) 880 struct mbuf *m; 881 int how; 882 { 883 struct mbuf *top, *n, *o; 884 885 MGET(n, how, m->m_type); 886 top = n; 887 if (!n) 888 goto nospace; 889 890 if (!m_dup_pkthdr(n, m, how)) 891 goto nospace; 892 n->m_len = m->m_len; 893 if (m->m_flags & M_EXT) { 894 n->m_data = m->m_data; 895 if (m->m_ext.ext_ref == NULL) 896 atomic_add_char(&mclrefcnt[mtocl(m->m_ext.ext_buf)], 1); 897 else { 898 int s = splimp(); 899 900 (*m->m_ext.ext_ref)(m->m_ext.ext_buf, 901 m->m_ext.ext_size); 902 splx(s); 903 } 904 n->m_ext = m->m_ext; 905 n->m_flags |= M_EXT; 906 } else { 907 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat ); 908 bcopy(mtod(m, char *), mtod(n, char *), n->m_len); 909 } 910 911 m = m->m_next; 912 while (m) { 913 MGET(o, how, m->m_type); 914 if (!o) 915 goto nospace; 916 917 n->m_next = o; 918 n = n->m_next; 919 920 n->m_len = m->m_len; 921 if (m->m_flags & M_EXT) { 922 n->m_data = m->m_data; 923 if (m->m_ext.ext_ref == NULL) { 924 atomic_add_char( 925 &mclrefcnt[mtocl(m->m_ext.ext_buf)], 1); 926 } else { 927 int s = splimp(); 928 929 (*m->m_ext.ext_ref)(m->m_ext.ext_buf, 930 m->m_ext.ext_size); 931 splx(s); 932 } 933 n->m_ext = m->m_ext; 934 n->m_flags |= M_EXT; 935 } else { 936 bcopy(mtod(m, char *), mtod(n, char *), n->m_len); 937 } 938 939 m = m->m_next; 940 } 941 return top; 942 nospace: 943 m_freem(top); 944 MCFail++; 945 return 0; 946 } 947 948 /* 949 * Copy data from an mbuf chain starting "off" bytes from the beginning, 950 * continuing for "len" bytes, into the indicated buffer. 951 */ 952 void 953 m_copydata(m, off, len, cp) 954 register struct mbuf *m; 955 register int off; 956 register int len; 957 caddr_t cp; 958 { 959 register unsigned count; 960 961 KASSERT(off >= 0, ("m_copydata, negative off %d", off)); 962 KASSERT(len >= 0, ("m_copydata, negative len %d", len)); 963 while (off > 0) { 964 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain")); 965 if (off < m->m_len) 966 break; 967 off -= m->m_len; 968 m = m->m_next; 969 } 970 while (len > 0) { 971 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain")); 972 count = min(m->m_len - off, len); 973 bcopy(mtod(m, caddr_t) + off, cp, count); 974 len -= count; 975 cp += count; 976 off = 0; 977 m = m->m_next; 978 } 979 } 980 981 /* 982 * Copy a packet header mbuf chain into a completely new chain, including 983 * copying any mbuf clusters. Use this instead of m_copypacket() when 984 * you need a writable copy of an mbuf chain. 985 */ 986 struct mbuf * 987 m_dup(m, how) 988 struct mbuf *m; 989 int how; 990 { 991 struct mbuf **p, *top = NULL; 992 int remain, moff, nsize; 993 994 /* Sanity check */ 995 if (m == NULL) 996 return (0); 997 KASSERT((m->m_flags & M_PKTHDR) != 0, ("%s: !PKTHDR", __FUNCTION__)); 998 999 /* While there's more data, get a new mbuf, tack it on, and fill it */ 1000 remain = m->m_pkthdr.len; 1001 moff = 0; 1002 p = ⊤ 1003 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */ 1004 struct mbuf *n; 1005 1006 /* Get the next new mbuf */ 1007 MGET(n, how, m->m_type); 1008 if (n == NULL) 1009 goto nospace; 1010 if (top == NULL) { /* first one, must be PKTHDR */ 1011 if (!m_dup_pkthdr(n, m, how)) 1012 goto nospace; 1013 nsize = MHLEN; 1014 } else /* not the first one */ 1015 nsize = MLEN; 1016 if (remain >= MINCLSIZE) { 1017 MCLGET(n, how); 1018 if ((n->m_flags & M_EXT) == 0) { 1019 (void)m_free(n); 1020 goto nospace; 1021 } 1022 nsize = MCLBYTES; 1023 } 1024 n->m_len = 0; 1025 1026 /* Link it into the new chain */ 1027 *p = n; 1028 p = &n->m_next; 1029 1030 /* Copy data from original mbuf(s) into new mbuf */ 1031 while (n->m_len < nsize && m != NULL) { 1032 int chunk = min(nsize - n->m_len, m->m_len - moff); 1033 1034 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk); 1035 moff += chunk; 1036 n->m_len += chunk; 1037 remain -= chunk; 1038 if (moff == m->m_len) { 1039 m = m->m_next; 1040 moff = 0; 1041 } 1042 } 1043 1044 /* Check correct total mbuf length */ 1045 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL), 1046 ("%s: bogus m_pkthdr.len", __FUNCTION__)); 1047 } 1048 return (top); 1049 1050 nospace: 1051 m_freem(top); 1052 MCFail++; 1053 return (0); 1054 } 1055 1056 /* 1057 * Concatenate mbuf chain n to m. 1058 * Both chains must be of the same type (e.g. MT_DATA). 1059 * Any m_pkthdr is not updated. 1060 */ 1061 void 1062 m_cat(m, n) 1063 register struct mbuf *m, *n; 1064 { 1065 while (m->m_next) 1066 m = m->m_next; 1067 while (n) { 1068 if (m->m_flags & M_EXT || 1069 m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) { 1070 /* just join the two chains */ 1071 m->m_next = n; 1072 return; 1073 } 1074 /* splat the data from one into the other */ 1075 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 1076 (u_int)n->m_len); 1077 m->m_len += n->m_len; 1078 n = m_free(n); 1079 } 1080 } 1081 1082 void 1083 m_adj(mp, req_len) 1084 struct mbuf *mp; 1085 int req_len; 1086 { 1087 register int len = req_len; 1088 register struct mbuf *m; 1089 register int count; 1090 1091 if ((m = mp) == NULL) 1092 return; 1093 if (len >= 0) { 1094 /* 1095 * Trim from head. 1096 */ 1097 while (m != NULL && len > 0) { 1098 if (m->m_len <= len) { 1099 len -= m->m_len; 1100 m->m_len = 0; 1101 m = m->m_next; 1102 } else { 1103 m->m_len -= len; 1104 m->m_data += len; 1105 len = 0; 1106 } 1107 } 1108 m = mp; 1109 if (mp->m_flags & M_PKTHDR) 1110 m->m_pkthdr.len -= (req_len - len); 1111 } else { 1112 /* 1113 * Trim from tail. Scan the mbuf chain, 1114 * calculating its length and finding the last mbuf. 1115 * If the adjustment only affects this mbuf, then just 1116 * adjust and return. Otherwise, rescan and truncate 1117 * after the remaining size. 1118 */ 1119 len = -len; 1120 count = 0; 1121 for (;;) { 1122 count += m->m_len; 1123 if (m->m_next == (struct mbuf *)0) 1124 break; 1125 m = m->m_next; 1126 } 1127 if (m->m_len >= len) { 1128 m->m_len -= len; 1129 if (mp->m_flags & M_PKTHDR) 1130 mp->m_pkthdr.len -= len; 1131 return; 1132 } 1133 count -= len; 1134 if (count < 0) 1135 count = 0; 1136 /* 1137 * Correct length for chain is "count". 1138 * Find the mbuf with last data, adjust its length, 1139 * and toss data from remaining mbufs on chain. 1140 */ 1141 m = mp; 1142 if (m->m_flags & M_PKTHDR) 1143 m->m_pkthdr.len = count; 1144 for (; m; m = m->m_next) { 1145 if (m->m_len >= count) { 1146 m->m_len = count; 1147 break; 1148 } 1149 count -= m->m_len; 1150 } 1151 while (m->m_next) 1152 (m = m->m_next) ->m_len = 0; 1153 } 1154 } 1155 1156 /* 1157 * Rearange an mbuf chain so that len bytes are contiguous 1158 * and in the data area of an mbuf (so that mtod and dtom 1159 * will work for a structure of size len). Returns the resulting 1160 * mbuf chain on success, frees it and returns null on failure. 1161 * If there is room, it will add up to max_protohdr-len extra bytes to the 1162 * contiguous region in an attempt to avoid being called next time. 1163 */ 1164 #define MPFail (mbstat.m_mpfail) 1165 1166 struct mbuf * 1167 m_pullup(n, len) 1168 register struct mbuf *n; 1169 int len; 1170 { 1171 register struct mbuf *m; 1172 register int count; 1173 int space; 1174 1175 /* 1176 * If first mbuf has no cluster, and has room for len bytes 1177 * without shifting current data, pullup into it, 1178 * otherwise allocate a new mbuf to prepend to the chain. 1179 */ 1180 if ((n->m_flags & M_EXT) == 0 && 1181 n->m_data + len < &n->m_dat[MLEN] && n->m_next) { 1182 if (n->m_len >= len) 1183 return (n); 1184 m = n; 1185 n = n->m_next; 1186 len -= m->m_len; 1187 } else { 1188 if (len > MHLEN) 1189 goto bad; 1190 MGET(m, M_DONTWAIT, n->m_type); 1191 if (m == 0) 1192 goto bad; 1193 m->m_len = 0; 1194 if (n->m_flags & M_PKTHDR) 1195 M_MOVE_PKTHDR(m, n); 1196 } 1197 space = &m->m_dat[MLEN] - (m->m_data + m->m_len); 1198 do { 1199 count = min(min(max(len, max_protohdr), space), n->m_len); 1200 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 1201 (unsigned)count); 1202 len -= count; 1203 m->m_len += count; 1204 n->m_len -= count; 1205 space -= count; 1206 if (n->m_len) 1207 n->m_data += count; 1208 else 1209 n = m_free(n); 1210 } while (len > 0 && n); 1211 if (len > 0) { 1212 (void) m_free(m); 1213 goto bad; 1214 } 1215 m->m_next = n; 1216 return (m); 1217 bad: 1218 m_freem(n); 1219 MPFail++; 1220 return (0); 1221 } 1222 1223 /* 1224 * Partition an mbuf chain in two pieces, returning the tail -- 1225 * all but the first len0 bytes. In case of failure, it returns NULL and 1226 * attempts to restore the chain to its original state. 1227 * 1228 * Note that the resulting mbufs might be read-only, because the new 1229 * mbuf can end up sharing an mbuf cluster with the original mbuf if 1230 * the "breaking point" happens to lie within a cluster mbuf. Use the 1231 * M_WRITABLE() macro to check for this case. 1232 */ 1233 struct mbuf * 1234 m_split(m0, len0, wait) 1235 register struct mbuf *m0; 1236 int len0, wait; 1237 { 1238 register struct mbuf *m, *n; 1239 unsigned len = len0, remain; 1240 1241 for (m = m0; m && len > m->m_len; m = m->m_next) 1242 len -= m->m_len; 1243 if (m == 0) 1244 return (0); 1245 remain = m->m_len - len; 1246 if (m0->m_flags & M_PKTHDR) { 1247 MGETHDR(n, wait, m0->m_type); 1248 if (n == 0) 1249 return (0); 1250 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif; 1251 n->m_pkthdr.len = m0->m_pkthdr.len - len0; 1252 m0->m_pkthdr.len = len0; 1253 if (m->m_flags & M_EXT) 1254 goto extpacket; 1255 if (remain > MHLEN) { 1256 /* m can't be the lead packet */ 1257 MH_ALIGN(n, 0); 1258 n->m_next = m_split(m, len, wait); 1259 if (n->m_next == 0) { 1260 (void) m_free(n); 1261 return (0); 1262 } else { 1263 n->m_len = 0; 1264 return (n); 1265 } 1266 } else 1267 MH_ALIGN(n, remain); 1268 } else if (remain == 0) { 1269 n = m->m_next; 1270 m->m_next = 0; 1271 return (n); 1272 } else { 1273 MGET(n, wait, m->m_type); 1274 if (n == 0) 1275 return (0); 1276 M_ALIGN(n, remain); 1277 } 1278 extpacket: 1279 if (m->m_flags & M_EXT) { 1280 n->m_flags |= M_EXT; 1281 n->m_ext = m->m_ext; 1282 if (m->m_ext.ext_ref == NULL) 1283 atomic_add_char(&mclrefcnt[mtocl(m->m_ext.ext_buf)], 1); 1284 else { 1285 int s = splimp(); 1286 1287 (*m->m_ext.ext_ref)(m->m_ext.ext_buf, 1288 m->m_ext.ext_size); 1289 splx(s); 1290 } 1291 n->m_data = m->m_data + len; 1292 } else { 1293 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain); 1294 } 1295 n->m_len = remain; 1296 m->m_len = len; 1297 n->m_next = m->m_next; 1298 m->m_next = 0; 1299 return (n); 1300 } 1301 /* 1302 * Routine to copy from device local memory into mbufs. 1303 */ 1304 struct mbuf * 1305 m_devget(buf, totlen, off0, ifp, copy) 1306 char *buf; 1307 int totlen, off0; 1308 struct ifnet *ifp; 1309 void (*copy) __P((char *from, caddr_t to, u_int len)); 1310 { 1311 register struct mbuf *m; 1312 struct mbuf *top = 0, **mp = ⊤ 1313 register int off = off0, len; 1314 register char *cp; 1315 char *epkt; 1316 1317 cp = buf; 1318 epkt = cp + totlen; 1319 if (off) { 1320 cp += off + 2 * sizeof(u_short); 1321 totlen -= 2 * sizeof(u_short); 1322 } 1323 MGETHDR(m, M_DONTWAIT, MT_DATA); 1324 if (m == 0) 1325 return (0); 1326 m->m_pkthdr.rcvif = ifp; 1327 m->m_pkthdr.len = totlen; 1328 m->m_len = MHLEN; 1329 1330 while (totlen > 0) { 1331 if (top) { 1332 MGET(m, M_DONTWAIT, MT_DATA); 1333 if (m == 0) { 1334 m_freem(top); 1335 return (0); 1336 } 1337 m->m_len = MLEN; 1338 } 1339 len = min(totlen, epkt - cp); 1340 if (len >= MINCLSIZE) { 1341 MCLGET(m, M_DONTWAIT); 1342 if (m->m_flags & M_EXT) 1343 m->m_len = len = min(len, MCLBYTES); 1344 else 1345 len = m->m_len; 1346 } else { 1347 /* 1348 * Place initial small packet/header at end of mbuf. 1349 */ 1350 if (len < m->m_len) { 1351 if (top == 0 && len + max_linkhdr <= m->m_len) 1352 m->m_data += max_linkhdr; 1353 m->m_len = len; 1354 } else 1355 len = m->m_len; 1356 } 1357 if (copy) 1358 copy(cp, mtod(m, caddr_t), (unsigned)len); 1359 else 1360 bcopy(cp, mtod(m, caddr_t), (unsigned)len); 1361 cp += len; 1362 *mp = m; 1363 mp = &m->m_next; 1364 totlen -= len; 1365 if (cp == epkt) 1366 cp = buf; 1367 } 1368 return (top); 1369 } 1370 1371 /* 1372 * Copy data from a buffer back into the indicated mbuf chain, 1373 * starting "off" bytes from the beginning, extending the mbuf 1374 * chain if necessary. 1375 */ 1376 void 1377 m_copyback(m0, off, len, cp) 1378 struct mbuf *m0; 1379 register int off; 1380 register int len; 1381 caddr_t cp; 1382 { 1383 register int mlen; 1384 register struct mbuf *m = m0, *n; 1385 int totlen = 0; 1386 1387 if (m0 == 0) 1388 return; 1389 while (off > (mlen = m->m_len)) { 1390 off -= mlen; 1391 totlen += mlen; 1392 if (m->m_next == 0) { 1393 n = m_getclr(M_DONTWAIT, m->m_type); 1394 if (n == 0) 1395 goto out; 1396 n->m_len = min(MLEN, len + off); 1397 m->m_next = n; 1398 } 1399 m = m->m_next; 1400 } 1401 while (len > 0) { 1402 mlen = min (m->m_len - off, len); 1403 bcopy(cp, off + mtod(m, caddr_t), (unsigned)mlen); 1404 cp += mlen; 1405 len -= mlen; 1406 mlen += off; 1407 off = 0; 1408 totlen += mlen; 1409 if (len == 0) 1410 break; 1411 if (m->m_next == 0) { 1412 n = m_get(M_DONTWAIT, m->m_type); 1413 if (n == 0) 1414 break; 1415 n->m_len = min(MLEN, len); 1416 m->m_next = n; 1417 } 1418 m = m->m_next; 1419 } 1420 out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen)) 1421 m->m_pkthdr.len = totlen; 1422 } 1423 1424 void 1425 m_print(const struct mbuf *m) 1426 { 1427 int len; 1428 const struct mbuf *m2; 1429 1430 len = m->m_pkthdr.len; 1431 m2 = m; 1432 while (len) { 1433 printf("%p %*D\n", m2, m2->m_len, (u_char *)m2->m_data, "-"); 1434 len -= m2->m_len; 1435 m2 = m2->m_next; 1436 } 1437 return; 1438 } 1439 1440 /* 1441 * "Move" mbuf pkthdr from "from" to "to". 1442 * "from" must have M_PKTHDR set, and "to" must be empty. 1443 */ 1444 void 1445 m_move_pkthdr(struct mbuf *to, struct mbuf *from) 1446 { 1447 KASSERT((to->m_flags & M_EXT) == 0, ("m_move_pkthdr: to has cluster")); 1448 1449 to->m_flags = from->m_flags & M_COPYFLAGS; 1450 to->m_data = to->m_pktdat; 1451 to->m_pkthdr = from->m_pkthdr; /* especially tags */ 1452 SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */ 1453 from->m_flags &= ~M_PKTHDR; 1454 } 1455 1456 /* 1457 * Duplicate "from"'s mbuf pkthdr in "to". 1458 * "from" must have M_PKTHDR set, and "to" must be empty. 1459 * In particular, this does a deep copy of the packet tags. 1460 */ 1461 int 1462 m_dup_pkthdr(struct mbuf *to, struct mbuf *from, int how) 1463 { 1464 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT); 1465 if ((to->m_flags & M_EXT) == 0) 1466 to->m_data = to->m_pktdat; 1467 to->m_pkthdr = from->m_pkthdr; 1468 SLIST_INIT(&to->m_pkthdr.tags); 1469 return (m_tag_copy_chain(to, from, how)); 1470 } 1471 1472 /* 1473 * Defragment a mbuf chain, returning the shortest possible 1474 * chain of mbufs and clusters. If allocation fails and 1475 * this cannot be completed, NULL will be returned, but 1476 * the passed in chain will be unchanged. Upon success, 1477 * the original chain will be freed, and the new chain 1478 * will be returned. 1479 * 1480 * If a non-packet header is passed in, the original 1481 * mbuf (chain?) will be returned unharmed. 1482 */ 1483 struct mbuf * 1484 m_defrag(struct mbuf *m0, int how) 1485 { 1486 struct mbuf *m_new = NULL, *m_final = NULL; 1487 int progress = 0, length; 1488 1489 if (!(m0->m_flags & M_PKTHDR)) 1490 return (m0); 1491 1492 #ifdef MBUF_STRESS_TEST 1493 if (m_defragrandomfailures) { 1494 int temp = arc4random() & 0xff; 1495 if (temp == 0xba) 1496 goto nospace; 1497 } 1498 #endif 1499 1500 if (m0->m_pkthdr.len > MHLEN) 1501 m_final = m_getcl(how, MT_DATA, M_PKTHDR); 1502 else 1503 m_final = m_gethdr(how, MT_DATA); 1504 1505 if (m_final == NULL) 1506 goto nospace; 1507 1508 if (m_dup_pkthdr(m_final, m0, how) == NULL) 1509 goto nospace; 1510 1511 m_new = m_final; 1512 1513 while (progress < m0->m_pkthdr.len) { 1514 length = m0->m_pkthdr.len - progress; 1515 if (length > MCLBYTES) 1516 length = MCLBYTES; 1517 1518 if (m_new == NULL) { 1519 if (length > MLEN) 1520 m_new = m_getcl(how, MT_DATA, 0); 1521 else 1522 m_new = m_get(how, MT_DATA); 1523 if (m_new == NULL) 1524 goto nospace; 1525 } 1526 1527 m_copydata(m0, progress, length, mtod(m_new, caddr_t)); 1528 progress += length; 1529 m_new->m_len = length; 1530 if (m_new != m_final) 1531 m_cat(m_final, m_new); 1532 m_new = NULL; 1533 } 1534 if (m0->m_next == NULL) 1535 m_defraguseless++; 1536 m_freem(m0); 1537 m0 = m_final; 1538 m_defragpackets++; 1539 m_defragbytes += m0->m_pkthdr.len; 1540 return (m0); 1541 nospace: 1542 m_defragfailure++; 1543 if (m_new) 1544 m_free(m_new); 1545 if (m_final) 1546 m_freem(m_final); 1547 return (NULL); 1548 } 1549