1 /* $NetBSD: uipc_mbuf.c,v 1.146 2012/04/29 16:36:53 dsl Exp $ */ 2 3 /*- 4 * Copyright (c) 1999, 2001 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility, 9 * NASA Ames Research Center. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 30 * POSSIBILITY OF SUCH DAMAGE. 31 */ 32 33 /* 34 * Copyright (c) 1982, 1986, 1988, 1991, 1993 35 * The Regents of the University of California. All rights reserved. 36 * 37 * Redistribution and use in source and binary forms, with or without 38 * modification, are permitted provided that the following conditions 39 * are met: 40 * 1. Redistributions of source code must retain the above copyright 41 * notice, this list of conditions and the following disclaimer. 42 * 2. Redistributions in binary form must reproduce the above copyright 43 * notice, this list of conditions and the following disclaimer in the 44 * documentation and/or other materials provided with the distribution. 45 * 3. Neither the name of the University nor the names of its contributors 46 * may be used to endorse or promote products derived from this software 47 * without specific prior written permission. 48 * 49 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 50 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 51 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 52 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 53 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 54 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 55 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 56 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 57 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 58 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 59 * SUCH DAMAGE. 60 * 61 * @(#)uipc_mbuf.c 8.4 (Berkeley) 2/14/95 62 */ 63 64 #include <sys/cdefs.h> 65 __KERNEL_RCSID(0, "$NetBSD: uipc_mbuf.c,v 1.146 2012/04/29 16:36:53 dsl Exp $"); 66 67 #include "opt_mbuftrace.h" 68 #include "opt_nmbclusters.h" 69 #include "opt_ddb.h" 70 71 #include <sys/param.h> 72 #include <sys/systm.h> 73 #include <sys/atomic.h> 74 #include <sys/cpu.h> 75 #include <sys/proc.h> 76 #include <sys/mbuf.h> 77 #include <sys/kernel.h> 78 #include <sys/syslog.h> 79 #include <sys/domain.h> 80 #include <sys/protosw.h> 81 #include <sys/percpu.h> 82 #include <sys/pool.h> 83 #include <sys/socket.h> 84 #include <sys/sysctl.h> 85 86 #include <net/if.h> 87 88 pool_cache_t mb_cache; /* mbuf cache */ 89 pool_cache_t mcl_cache; /* mbuf cluster cache */ 90 91 struct mbstat mbstat; 92 int max_linkhdr; 93 int max_protohdr; 94 int max_hdr; 95 int max_datalen; 96 97 static int mb_ctor(void *, void *, int); 98 99 static void sysctl_kern_mbuf_setup(void); 100 101 static struct sysctllog *mbuf_sysctllog; 102 103 static struct mbuf *m_copym0(struct mbuf *, int, int, int, int); 104 static struct mbuf *m_split0(struct mbuf *, int, int, int); 105 static int m_copyback0(struct mbuf **, int, int, const void *, int, int); 106 107 /* flags for m_copyback0 */ 108 #define M_COPYBACK0_COPYBACK 0x0001 /* copyback from cp */ 109 #define M_COPYBACK0_PRESERVE 0x0002 /* preserve original data */ 110 #define M_COPYBACK0_COW 0x0004 /* do copy-on-write */ 111 #define M_COPYBACK0_EXTEND 0x0008 /* extend chain */ 112 113 static const char mclpool_warnmsg[] = 114 "WARNING: mclpool limit reached; increase kern.mbuf.nmbclusters"; 115 116 MALLOC_DEFINE(M_MBUF, "mbuf", "mbuf"); 117 118 static percpu_t *mbstat_percpu; 119 120 #ifdef MBUFTRACE 121 struct mownerhead mowners = LIST_HEAD_INITIALIZER(mowners); 122 struct mowner unknown_mowners[] = { 123 MOWNER_INIT("unknown", "free"), 124 MOWNER_INIT("unknown", "data"), 125 MOWNER_INIT("unknown", "header"), 126 MOWNER_INIT("unknown", "soname"), 127 MOWNER_INIT("unknown", "soopts"), 128 MOWNER_INIT("unknown", "ftable"), 129 MOWNER_INIT("unknown", "control"), 130 MOWNER_INIT("unknown", "oobdata"), 131 }; 132 struct mowner revoked_mowner = MOWNER_INIT("revoked", ""); 133 #endif 134 135 #define MEXT_ISEMBEDDED(m) ((m)->m_ext_ref == (m)) 136 137 #define MCLADDREFERENCE(o, n) \ 138 do { \ 139 KASSERT(((o)->m_flags & M_EXT) != 0); \ 140 KASSERT(((n)->m_flags & M_EXT) == 0); \ 141 KASSERT((o)->m_ext.ext_refcnt >= 1); \ 142 (n)->m_flags |= ((o)->m_flags & M_EXTCOPYFLAGS); \ 143 atomic_inc_uint(&(o)->m_ext.ext_refcnt); \ 144 (n)->m_ext_ref = (o)->m_ext_ref; \ 145 mowner_ref((n), (n)->m_flags); \ 146 MCLREFDEBUGN((n), __FILE__, __LINE__); \ 147 } while (/* CONSTCOND */ 0) 148 149 static int 150 nmbclusters_limit(void) 151 { 152 #if defined(PMAP_MAP_POOLPAGE) 153 /* direct mapping, doesn't use space in kmem_map */ 154 vsize_t max_size = physmem / 4; 155 #else 156 vsize_t max_size = MIN(physmem / 4, nkmempages / 4); 157 #endif 158 159 max_size = max_size * PAGE_SIZE / MCLBYTES; 160 #ifdef NMBCLUSTERS_MAX 161 max_size = MIN(max_size, NMBCLUSTERS_MAX); 162 #endif 163 164 #ifdef NMBCLUSTERS 165 return MIN(max_size, NMBCLUSTERS); 166 #else 167 return max_size; 168 #endif 169 } 170 171 /* 172 * Initialize the mbuf allocator. 173 */ 174 void 175 mbinit(void) 176 { 177 178 CTASSERT(sizeof(struct _m_ext) <= MHLEN); 179 CTASSERT(sizeof(struct mbuf) == MSIZE); 180 181 sysctl_kern_mbuf_setup(); 182 183 mb_cache = pool_cache_init(msize, 0, 0, 0, "mbpl", 184 NULL, IPL_VM, mb_ctor, NULL, NULL); 185 KASSERT(mb_cache != NULL); 186 187 mcl_cache = pool_cache_init(mclbytes, 0, 0, 0, "mclpl", NULL, 188 IPL_VM, NULL, NULL, NULL); 189 KASSERT(mcl_cache != NULL); 190 191 pool_cache_set_drain_hook(mb_cache, m_reclaim, NULL); 192 pool_cache_set_drain_hook(mcl_cache, m_reclaim, NULL); 193 194 /* 195 * Set an arbitrary default limit on the number of mbuf clusters. 196 */ 197 #ifdef NMBCLUSTERS 198 nmbclusters = nmbclusters_limit(); 199 #else 200 nmbclusters = MAX(1024, 201 (vsize_t)physmem * PAGE_SIZE / MCLBYTES / 16); 202 nmbclusters = MIN(nmbclusters, nmbclusters_limit()); 203 #endif 204 205 /* 206 * Set the hard limit on the mclpool to the number of 207 * mbuf clusters the kernel is to support. Log the limit 208 * reached message max once a minute. 209 */ 210 pool_cache_sethardlimit(mcl_cache, nmbclusters, mclpool_warnmsg, 60); 211 212 mbstat_percpu = percpu_alloc(sizeof(struct mbstat_cpu)); 213 214 /* 215 * Set a low water mark for both mbufs and clusters. This should 216 * help ensure that they can be allocated in a memory starvation 217 * situation. This is important for e.g. diskless systems which 218 * must allocate mbufs in order for the pagedaemon to clean pages. 219 */ 220 pool_cache_setlowat(mb_cache, mblowat); 221 pool_cache_setlowat(mcl_cache, mcllowat); 222 223 #ifdef MBUFTRACE 224 { 225 /* 226 * Attach the unknown mowners. 227 */ 228 int i; 229 MOWNER_ATTACH(&revoked_mowner); 230 for (i = sizeof(unknown_mowners)/sizeof(unknown_mowners[0]); 231 i-- > 0; ) 232 MOWNER_ATTACH(&unknown_mowners[i]); 233 } 234 #endif 235 } 236 237 /* 238 * sysctl helper routine for the kern.mbuf subtree. 239 * nmbclusters, mblowat and mcllowat need range 240 * checking and pool tweaking after being reset. 241 */ 242 static int 243 sysctl_kern_mbuf(SYSCTLFN_ARGS) 244 { 245 int error, newval; 246 struct sysctlnode node; 247 248 node = *rnode; 249 node.sysctl_data = &newval; 250 switch (rnode->sysctl_num) { 251 case MBUF_NMBCLUSTERS: 252 case MBUF_MBLOWAT: 253 case MBUF_MCLLOWAT: 254 newval = *(int*)rnode->sysctl_data; 255 break; 256 default: 257 return (EOPNOTSUPP); 258 } 259 260 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 261 if (error || newp == NULL) 262 return (error); 263 if (newval < 0) 264 return (EINVAL); 265 266 switch (node.sysctl_num) { 267 case MBUF_NMBCLUSTERS: 268 if (newval < nmbclusters) 269 return (EINVAL); 270 if (newval > nmbclusters_limit()) 271 return (EINVAL); 272 nmbclusters = newval; 273 pool_cache_sethardlimit(mcl_cache, nmbclusters, 274 mclpool_warnmsg, 60); 275 break; 276 case MBUF_MBLOWAT: 277 mblowat = newval; 278 pool_cache_setlowat(mb_cache, mblowat); 279 break; 280 case MBUF_MCLLOWAT: 281 mcllowat = newval; 282 pool_cache_setlowat(mcl_cache, mcllowat); 283 break; 284 } 285 286 return (0); 287 } 288 289 #ifdef MBUFTRACE 290 static void 291 mowner_conver_to_user_cb(void *v1, void *v2, struct cpu_info *ci) 292 { 293 struct mowner_counter *mc = v1; 294 struct mowner_user *mo_user = v2; 295 int i; 296 297 for (i = 0; i < MOWNER_COUNTER_NCOUNTERS; i++) { 298 mo_user->mo_counter[i] += mc->mc_counter[i]; 299 } 300 } 301 302 static void 303 mowner_convert_to_user(struct mowner *mo, struct mowner_user *mo_user) 304 { 305 306 memset(mo_user, 0, sizeof(*mo_user)); 307 CTASSERT(sizeof(mo_user->mo_name) == sizeof(mo->mo_name)); 308 CTASSERT(sizeof(mo_user->mo_descr) == sizeof(mo->mo_descr)); 309 memcpy(mo_user->mo_name, mo->mo_name, sizeof(mo->mo_name)); 310 memcpy(mo_user->mo_descr, mo->mo_descr, sizeof(mo->mo_descr)); 311 percpu_foreach(mo->mo_counters, mowner_conver_to_user_cb, mo_user); 312 } 313 314 static int 315 sysctl_kern_mbuf_mowners(SYSCTLFN_ARGS) 316 { 317 struct mowner *mo; 318 size_t len = 0; 319 int error = 0; 320 321 if (namelen != 0) 322 return (EINVAL); 323 if (newp != NULL) 324 return (EPERM); 325 326 LIST_FOREACH(mo, &mowners, mo_link) { 327 struct mowner_user mo_user; 328 329 mowner_convert_to_user(mo, &mo_user); 330 331 if (oldp != NULL) { 332 if (*oldlenp - len < sizeof(mo_user)) { 333 error = ENOMEM; 334 break; 335 } 336 error = copyout(&mo_user, (char *)oldp + len, 337 sizeof(mo_user)); 338 if (error) 339 break; 340 } 341 len += sizeof(mo_user); 342 } 343 344 if (error == 0) 345 *oldlenp = len; 346 347 return (error); 348 } 349 #endif /* MBUFTRACE */ 350 351 static void 352 mbstat_conver_to_user_cb(void *v1, void *v2, struct cpu_info *ci) 353 { 354 struct mbstat_cpu *mbsc = v1; 355 struct mbstat *mbs = v2; 356 int i; 357 358 for (i = 0; i < __arraycount(mbs->m_mtypes); i++) { 359 mbs->m_mtypes[i] += mbsc->m_mtypes[i]; 360 } 361 } 362 363 static void 364 mbstat_convert_to_user(struct mbstat *mbs) 365 { 366 367 memset(mbs, 0, sizeof(*mbs)); 368 mbs->m_drain = mbstat.m_drain; 369 percpu_foreach(mbstat_percpu, mbstat_conver_to_user_cb, mbs); 370 } 371 372 static int 373 sysctl_kern_mbuf_stats(SYSCTLFN_ARGS) 374 { 375 struct sysctlnode node; 376 struct mbstat mbs; 377 378 mbstat_convert_to_user(&mbs); 379 node = *rnode; 380 node.sysctl_data = &mbs; 381 node.sysctl_size = sizeof(mbs); 382 return sysctl_lookup(SYSCTLFN_CALL(&node)); 383 } 384 385 static void 386 sysctl_kern_mbuf_setup(void) 387 { 388 389 KASSERT(mbuf_sysctllog == NULL); 390 sysctl_createv(&mbuf_sysctllog, 0, NULL, NULL, 391 CTLFLAG_PERMANENT, 392 CTLTYPE_NODE, "kern", NULL, 393 NULL, 0, NULL, 0, 394 CTL_KERN, CTL_EOL); 395 sysctl_createv(&mbuf_sysctllog, 0, NULL, NULL, 396 CTLFLAG_PERMANENT, 397 CTLTYPE_NODE, "mbuf", 398 SYSCTL_DESCR("mbuf control variables"), 399 NULL, 0, NULL, 0, 400 CTL_KERN, KERN_MBUF, CTL_EOL); 401 402 sysctl_createv(&mbuf_sysctllog, 0, NULL, NULL, 403 CTLFLAG_PERMANENT|CTLFLAG_IMMEDIATE, 404 CTLTYPE_INT, "msize", 405 SYSCTL_DESCR("mbuf base size"), 406 NULL, msize, NULL, 0, 407 CTL_KERN, KERN_MBUF, MBUF_MSIZE, CTL_EOL); 408 sysctl_createv(&mbuf_sysctllog, 0, NULL, NULL, 409 CTLFLAG_PERMANENT|CTLFLAG_IMMEDIATE, 410 CTLTYPE_INT, "mclbytes", 411 SYSCTL_DESCR("mbuf cluster size"), 412 NULL, mclbytes, NULL, 0, 413 CTL_KERN, KERN_MBUF, MBUF_MCLBYTES, CTL_EOL); 414 sysctl_createv(&mbuf_sysctllog, 0, NULL, NULL, 415 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 416 CTLTYPE_INT, "nmbclusters", 417 SYSCTL_DESCR("Limit on the number of mbuf clusters"), 418 sysctl_kern_mbuf, 0, &nmbclusters, 0, 419 CTL_KERN, KERN_MBUF, MBUF_NMBCLUSTERS, CTL_EOL); 420 sysctl_createv(&mbuf_sysctllog, 0, NULL, NULL, 421 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 422 CTLTYPE_INT, "mblowat", 423 SYSCTL_DESCR("mbuf low water mark"), 424 sysctl_kern_mbuf, 0, &mblowat, 0, 425 CTL_KERN, KERN_MBUF, MBUF_MBLOWAT, CTL_EOL); 426 sysctl_createv(&mbuf_sysctllog, 0, NULL, NULL, 427 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 428 CTLTYPE_INT, "mcllowat", 429 SYSCTL_DESCR("mbuf cluster low water mark"), 430 sysctl_kern_mbuf, 0, &mcllowat, 0, 431 CTL_KERN, KERN_MBUF, MBUF_MCLLOWAT, CTL_EOL); 432 sysctl_createv(&mbuf_sysctllog, 0, NULL, NULL, 433 CTLFLAG_PERMANENT, 434 CTLTYPE_STRUCT, "stats", 435 SYSCTL_DESCR("mbuf allocation statistics"), 436 sysctl_kern_mbuf_stats, 0, NULL, 0, 437 CTL_KERN, KERN_MBUF, MBUF_STATS, CTL_EOL); 438 #ifdef MBUFTRACE 439 sysctl_createv(&mbuf_sysctllog, 0, NULL, NULL, 440 CTLFLAG_PERMANENT, 441 CTLTYPE_STRUCT, "mowners", 442 SYSCTL_DESCR("Information about mbuf owners"), 443 sysctl_kern_mbuf_mowners, 0, NULL, 0, 444 CTL_KERN, KERN_MBUF, MBUF_MOWNERS, CTL_EOL); 445 #endif /* MBUFTRACE */ 446 } 447 448 static int 449 mb_ctor(void *arg, void *object, int flags) 450 { 451 struct mbuf *m = object; 452 453 #ifdef POOL_VTOPHYS 454 m->m_paddr = POOL_VTOPHYS(m); 455 #else 456 m->m_paddr = M_PADDR_INVALID; 457 #endif 458 return (0); 459 } 460 461 void 462 m_reclaim(void *arg, int flags) 463 { 464 struct domain *dp; 465 const struct protosw *pr; 466 struct ifnet *ifp; 467 int s; 468 469 KERNEL_LOCK(1, NULL); 470 s = splvm(); 471 DOMAIN_FOREACH(dp) { 472 for (pr = dp->dom_protosw; 473 pr < dp->dom_protoswNPROTOSW; pr++) 474 if (pr->pr_drain) 475 (*pr->pr_drain)(); 476 } 477 IFNET_FOREACH(ifp) { 478 if (ifp->if_drain) 479 (*ifp->if_drain)(ifp); 480 } 481 splx(s); 482 mbstat.m_drain++; 483 KERNEL_UNLOCK_ONE(NULL); 484 } 485 486 /* 487 * Space allocation routines. 488 * These are also available as macros 489 * for critical paths. 490 */ 491 struct mbuf * 492 m_get(int nowait, int type) 493 { 494 struct mbuf *m; 495 496 KASSERT(type != MT_FREE); 497 498 m = pool_cache_get(mb_cache, 499 nowait == M_WAIT ? PR_WAITOK|PR_LIMITFAIL : 0); 500 if (m == NULL) 501 return NULL; 502 503 mbstat_type_add(type, 1); 504 mowner_init(m, type); 505 m->m_ext_ref = m; 506 m->m_type = type; 507 m->m_next = NULL; 508 m->m_nextpkt = NULL; 509 m->m_data = m->m_dat; 510 m->m_flags = 0; 511 512 return m; 513 } 514 515 struct mbuf * 516 m_gethdr(int nowait, int type) 517 { 518 struct mbuf *m; 519 520 m = m_get(nowait, type); 521 if (m == NULL) 522 return NULL; 523 524 m->m_data = m->m_pktdat; 525 m->m_flags = M_PKTHDR; 526 m->m_pkthdr.rcvif = NULL; 527 m->m_pkthdr.csum_flags = 0; 528 m->m_pkthdr.csum_data = 0; 529 SLIST_INIT(&m->m_pkthdr.tags); 530 531 return m; 532 } 533 534 struct mbuf * 535 m_getclr(int nowait, int type) 536 { 537 struct mbuf *m; 538 539 MGET(m, nowait, type); 540 if (m == 0) 541 return (NULL); 542 memset(mtod(m, void *), 0, MLEN); 543 return (m); 544 } 545 546 void 547 m_clget(struct mbuf *m, int nowait) 548 { 549 550 MCLGET(m, nowait); 551 } 552 553 struct mbuf * 554 m_free(struct mbuf *m) 555 { 556 struct mbuf *n; 557 558 MFREE(m, n); 559 return (n); 560 } 561 562 void 563 m_freem(struct mbuf *m) 564 { 565 struct mbuf *n; 566 567 if (m == NULL) 568 return; 569 do { 570 MFREE(m, n); 571 m = n; 572 } while (m); 573 } 574 575 #ifdef MBUFTRACE 576 /* 577 * Walk a chain of mbufs, claiming ownership of each mbuf in the chain. 578 */ 579 void 580 m_claimm(struct mbuf *m, struct mowner *mo) 581 { 582 583 for (; m != NULL; m = m->m_next) 584 MCLAIM(m, mo); 585 } 586 #endif 587 588 /* 589 * Mbuffer utility routines. 590 */ 591 592 /* 593 * Lesser-used path for M_PREPEND: 594 * allocate new mbuf to prepend to chain, 595 * copy junk along. 596 */ 597 struct mbuf * 598 m_prepend(struct mbuf *m, int len, int how) 599 { 600 struct mbuf *mn; 601 602 MGET(mn, how, m->m_type); 603 if (mn == NULL) { 604 m_freem(m); 605 return (NULL); 606 } 607 if (m->m_flags & M_PKTHDR) { 608 M_MOVE_PKTHDR(mn, m); 609 } else { 610 MCLAIM(mn, m->m_owner); 611 } 612 mn->m_next = m; 613 m = mn; 614 if (len < MHLEN) 615 MH_ALIGN(m, len); 616 m->m_len = len; 617 return (m); 618 } 619 620 /* 621 * Make a copy of an mbuf chain starting "off0" bytes from the beginning, 622 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf. 623 * The wait parameter is a choice of M_WAIT/M_DONTWAIT from caller. 624 */ 625 int MCFail; 626 627 struct mbuf * 628 m_copym(struct mbuf *m, int off0, int len, int wait) 629 { 630 631 return m_copym0(m, off0, len, wait, 0); /* shallow copy on M_EXT */ 632 } 633 634 struct mbuf * 635 m_dup(struct mbuf *m, int off0, int len, int wait) 636 { 637 638 return m_copym0(m, off0, len, wait, 1); /* deep copy */ 639 } 640 641 static struct mbuf * 642 m_copym0(struct mbuf *m, int off0, int len, int wait, int deep) 643 { 644 struct mbuf *n, **np; 645 int off = off0; 646 struct mbuf *top; 647 int copyhdr = 0; 648 649 if (off < 0 || len < 0) 650 panic("m_copym: off %d, len %d", off, len); 651 if (off == 0 && m->m_flags & M_PKTHDR) 652 copyhdr = 1; 653 while (off > 0) { 654 if (m == 0) 655 panic("m_copym: m == 0, off %d", off); 656 if (off < m->m_len) 657 break; 658 off -= m->m_len; 659 m = m->m_next; 660 } 661 np = ⊤ 662 top = 0; 663 while (len > 0) { 664 if (m == 0) { 665 if (len != M_COPYALL) 666 panic("m_copym: m == 0, len %d [!COPYALL]", 667 len); 668 break; 669 } 670 MGET(n, wait, m->m_type); 671 *np = n; 672 if (n == 0) 673 goto nospace; 674 MCLAIM(n, m->m_owner); 675 if (copyhdr) { 676 M_COPY_PKTHDR(n, m); 677 if (len == M_COPYALL) 678 n->m_pkthdr.len -= off0; 679 else 680 n->m_pkthdr.len = len; 681 copyhdr = 0; 682 } 683 n->m_len = min(len, m->m_len - off); 684 if (m->m_flags & M_EXT) { 685 if (!deep) { 686 n->m_data = m->m_data + off; 687 MCLADDREFERENCE(m, n); 688 } else { 689 /* 690 * we are unsure about the way m was allocated. 691 * copy into multiple MCLBYTES cluster mbufs. 692 */ 693 MCLGET(n, wait); 694 n->m_len = 0; 695 n->m_len = M_TRAILINGSPACE(n); 696 n->m_len = min(n->m_len, len); 697 n->m_len = min(n->m_len, m->m_len - off); 698 memcpy(mtod(n, void *), mtod(m, char *) + off, 699 (unsigned)n->m_len); 700 } 701 } else 702 memcpy(mtod(n, void *), mtod(m, char *) + off, 703 (unsigned)n->m_len); 704 if (len != M_COPYALL) 705 len -= n->m_len; 706 off += n->m_len; 707 #ifdef DIAGNOSTIC 708 if (off > m->m_len) 709 panic("m_copym0 overrun"); 710 #endif 711 if (off == m->m_len) { 712 m = m->m_next; 713 off = 0; 714 } 715 np = &n->m_next; 716 } 717 if (top == 0) 718 MCFail++; 719 return (top); 720 nospace: 721 m_freem(top); 722 MCFail++; 723 return (NULL); 724 } 725 726 /* 727 * Copy an entire packet, including header (which must be present). 728 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'. 729 */ 730 struct mbuf * 731 m_copypacket(struct mbuf *m, int how) 732 { 733 struct mbuf *top, *n, *o; 734 735 MGET(n, how, m->m_type); 736 top = n; 737 if (!n) 738 goto nospace; 739 740 MCLAIM(n, m->m_owner); 741 M_COPY_PKTHDR(n, m); 742 n->m_len = m->m_len; 743 if (m->m_flags & M_EXT) { 744 n->m_data = m->m_data; 745 MCLADDREFERENCE(m, n); 746 } else { 747 memcpy(mtod(n, char *), mtod(m, char *), n->m_len); 748 } 749 750 m = m->m_next; 751 while (m) { 752 MGET(o, how, m->m_type); 753 if (!o) 754 goto nospace; 755 756 MCLAIM(o, m->m_owner); 757 n->m_next = o; 758 n = n->m_next; 759 760 n->m_len = m->m_len; 761 if (m->m_flags & M_EXT) { 762 n->m_data = m->m_data; 763 MCLADDREFERENCE(m, n); 764 } else { 765 memcpy(mtod(n, char *), mtod(m, char *), n->m_len); 766 } 767 768 m = m->m_next; 769 } 770 return top; 771 nospace: 772 m_freem(top); 773 MCFail++; 774 return NULL; 775 } 776 777 /* 778 * Copy data from an mbuf chain starting "off" bytes from the beginning, 779 * continuing for "len" bytes, into the indicated buffer. 780 */ 781 void 782 m_copydata(struct mbuf *m, int off, int len, void *vp) 783 { 784 unsigned count; 785 void * cp = vp; 786 787 if (off < 0 || len < 0) 788 panic("m_copydata: off %d, len %d", off, len); 789 while (off > 0) { 790 if (m == NULL) 791 panic("m_copydata: m == NULL, off %d", off); 792 if (off < m->m_len) 793 break; 794 off -= m->m_len; 795 m = m->m_next; 796 } 797 while (len > 0) { 798 if (m == NULL) 799 panic("m_copydata: m == NULL, len %d", len); 800 count = min(m->m_len - off, len); 801 memcpy(cp, mtod(m, char *) + off, count); 802 len -= count; 803 cp = (char *)cp + count; 804 off = 0; 805 m = m->m_next; 806 } 807 } 808 809 /* 810 * Concatenate mbuf chain n to m. 811 * n might be copied into m (when n->m_len is small), therefore data portion of 812 * n could be copied into an mbuf of different mbuf type. 813 * Any m_pkthdr is not updated. 814 */ 815 void 816 m_cat(struct mbuf *m, struct mbuf *n) 817 { 818 819 while (m->m_next) 820 m = m->m_next; 821 while (n) { 822 if (M_READONLY(m) || n->m_len > M_TRAILINGSPACE(m)) { 823 /* just join the two chains */ 824 m->m_next = n; 825 return; 826 } 827 /* splat the data from one into the other */ 828 memcpy(mtod(m, char *) + m->m_len, mtod(n, void *), 829 (u_int)n->m_len); 830 m->m_len += n->m_len; 831 n = m_free(n); 832 } 833 } 834 835 void 836 m_adj(struct mbuf *mp, int req_len) 837 { 838 int len = req_len; 839 struct mbuf *m; 840 int count; 841 842 if ((m = mp) == NULL) 843 return; 844 if (len >= 0) { 845 /* 846 * Trim from head. 847 */ 848 while (m != NULL && len > 0) { 849 if (m->m_len <= len) { 850 len -= m->m_len; 851 m->m_len = 0; 852 m = m->m_next; 853 } else { 854 m->m_len -= len; 855 m->m_data += len; 856 len = 0; 857 } 858 } 859 m = mp; 860 if (mp->m_flags & M_PKTHDR) 861 m->m_pkthdr.len -= (req_len - len); 862 } else { 863 /* 864 * Trim from tail. Scan the mbuf chain, 865 * calculating its length and finding the last mbuf. 866 * If the adjustment only affects this mbuf, then just 867 * adjust and return. Otherwise, rescan and truncate 868 * after the remaining size. 869 */ 870 len = -len; 871 count = 0; 872 for (;;) { 873 count += m->m_len; 874 if (m->m_next == (struct mbuf *)0) 875 break; 876 m = m->m_next; 877 } 878 if (m->m_len >= len) { 879 m->m_len -= len; 880 if (mp->m_flags & M_PKTHDR) 881 mp->m_pkthdr.len -= len; 882 return; 883 } 884 count -= len; 885 if (count < 0) 886 count = 0; 887 /* 888 * Correct length for chain is "count". 889 * Find the mbuf with last data, adjust its length, 890 * and toss data from remaining mbufs on chain. 891 */ 892 m = mp; 893 if (m->m_flags & M_PKTHDR) 894 m->m_pkthdr.len = count; 895 for (; m; m = m->m_next) { 896 if (m->m_len >= count) { 897 m->m_len = count; 898 break; 899 } 900 count -= m->m_len; 901 } 902 if (m) 903 while (m->m_next) 904 (m = m->m_next)->m_len = 0; 905 } 906 } 907 908 /* 909 * Rearrange an mbuf chain so that len bytes are contiguous 910 * and in the data area of an mbuf (so that mtod and dtom 911 * will work for a structure of size len). Returns the resulting 912 * mbuf chain on success, frees it and returns null on failure. 913 * If there is room, it will add up to max_protohdr-len extra bytes to the 914 * contiguous region in an attempt to avoid being called next time. 915 */ 916 int MPFail; 917 918 struct mbuf * 919 m_pullup(struct mbuf *n, int len) 920 { 921 struct mbuf *m; 922 int count; 923 int space; 924 925 /* 926 * If first mbuf has no cluster, and has room for len bytes 927 * without shifting current data, pullup into it, 928 * otherwise allocate a new mbuf to prepend to the chain. 929 */ 930 if ((n->m_flags & M_EXT) == 0 && 931 n->m_data + len < &n->m_dat[MLEN] && n->m_next) { 932 if (n->m_len >= len) 933 return (n); 934 m = n; 935 n = n->m_next; 936 len -= m->m_len; 937 } else { 938 if (len > MHLEN) 939 goto bad; 940 MGET(m, M_DONTWAIT, n->m_type); 941 if (m == 0) 942 goto bad; 943 MCLAIM(m, n->m_owner); 944 m->m_len = 0; 945 if (n->m_flags & M_PKTHDR) { 946 M_MOVE_PKTHDR(m, n); 947 } 948 } 949 space = &m->m_dat[MLEN] - (m->m_data + m->m_len); 950 do { 951 count = min(min(max(len, max_protohdr), space), n->m_len); 952 memcpy(mtod(m, char *) + m->m_len, mtod(n, void *), 953 (unsigned)count); 954 len -= count; 955 m->m_len += count; 956 n->m_len -= count; 957 space -= count; 958 if (n->m_len) 959 n->m_data += count; 960 else 961 n = m_free(n); 962 } while (len > 0 && n); 963 if (len > 0) { 964 (void) m_free(m); 965 goto bad; 966 } 967 m->m_next = n; 968 return (m); 969 bad: 970 m_freem(n); 971 MPFail++; 972 return (NULL); 973 } 974 975 /* 976 * Like m_pullup(), except a new mbuf is always allocated, and we allow 977 * the amount of empty space before the data in the new mbuf to be specified 978 * (in the event that the caller expects to prepend later). 979 */ 980 int MSFail; 981 982 struct mbuf * 983 m_copyup(struct mbuf *n, int len, int dstoff) 984 { 985 struct mbuf *m; 986 int count, space; 987 988 if (len > (MHLEN - dstoff)) 989 goto bad; 990 MGET(m, M_DONTWAIT, n->m_type); 991 if (m == NULL) 992 goto bad; 993 MCLAIM(m, n->m_owner); 994 m->m_len = 0; 995 if (n->m_flags & M_PKTHDR) { 996 M_MOVE_PKTHDR(m, n); 997 } 998 m->m_data += dstoff; 999 space = &m->m_dat[MLEN] - (m->m_data + m->m_len); 1000 do { 1001 count = min(min(max(len, max_protohdr), space), n->m_len); 1002 memcpy(mtod(m, char *) + m->m_len, mtod(n, void *), 1003 (unsigned)count); 1004 len -= count; 1005 m->m_len += count; 1006 n->m_len -= count; 1007 space -= count; 1008 if (n->m_len) 1009 n->m_data += count; 1010 else 1011 n = m_free(n); 1012 } while (len > 0 && n); 1013 if (len > 0) { 1014 (void) m_free(m); 1015 goto bad; 1016 } 1017 m->m_next = n; 1018 return (m); 1019 bad: 1020 m_freem(n); 1021 MSFail++; 1022 return (NULL); 1023 } 1024 1025 /* 1026 * Partition an mbuf chain in two pieces, returning the tail -- 1027 * all but the first len0 bytes. In case of failure, it returns NULL and 1028 * attempts to restore the chain to its original state. 1029 */ 1030 struct mbuf * 1031 m_split(struct mbuf *m0, int len0, int wait) 1032 { 1033 1034 return m_split0(m0, len0, wait, 1); 1035 } 1036 1037 static struct mbuf * 1038 m_split0(struct mbuf *m0, int len0, int wait, int copyhdr) 1039 { 1040 struct mbuf *m, *n; 1041 unsigned len = len0, remain, len_save; 1042 1043 for (m = m0; m && len > m->m_len; m = m->m_next) 1044 len -= m->m_len; 1045 if (m == 0) 1046 return (NULL); 1047 remain = m->m_len - len; 1048 if (copyhdr && (m0->m_flags & M_PKTHDR)) { 1049 MGETHDR(n, wait, m0->m_type); 1050 if (n == 0) 1051 return (NULL); 1052 MCLAIM(n, m0->m_owner); 1053 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif; 1054 n->m_pkthdr.len = m0->m_pkthdr.len - len0; 1055 len_save = m0->m_pkthdr.len; 1056 m0->m_pkthdr.len = len0; 1057 if (m->m_flags & M_EXT) 1058 goto extpacket; 1059 if (remain > MHLEN) { 1060 /* m can't be the lead packet */ 1061 MH_ALIGN(n, 0); 1062 n->m_len = 0; 1063 n->m_next = m_split(m, len, wait); 1064 if (n->m_next == 0) { 1065 (void) m_free(n); 1066 m0->m_pkthdr.len = len_save; 1067 return (NULL); 1068 } else 1069 return (n); 1070 } else 1071 MH_ALIGN(n, remain); 1072 } else if (remain == 0) { 1073 n = m->m_next; 1074 m->m_next = 0; 1075 return (n); 1076 } else { 1077 MGET(n, wait, m->m_type); 1078 if (n == 0) 1079 return (NULL); 1080 MCLAIM(n, m->m_owner); 1081 M_ALIGN(n, remain); 1082 } 1083 extpacket: 1084 if (m->m_flags & M_EXT) { 1085 n->m_data = m->m_data + len; 1086 MCLADDREFERENCE(m, n); 1087 } else { 1088 memcpy(mtod(n, void *), mtod(m, char *) + len, remain); 1089 } 1090 n->m_len = remain; 1091 m->m_len = len; 1092 n->m_next = m->m_next; 1093 m->m_next = 0; 1094 return (n); 1095 } 1096 /* 1097 * Routine to copy from device local memory into mbufs. 1098 */ 1099 struct mbuf * 1100 m_devget(char *buf, int totlen, int off0, struct ifnet *ifp, 1101 void (*copy)(const void *from, void *to, size_t len)) 1102 { 1103 struct mbuf *m; 1104 struct mbuf *top = 0, **mp = ⊤ 1105 int off = off0, len; 1106 char *cp; 1107 char *epkt; 1108 1109 cp = buf; 1110 epkt = cp + totlen; 1111 if (off) { 1112 /* 1113 * If 'off' is non-zero, packet is trailer-encapsulated, 1114 * so we have to skip the type and length fields. 1115 */ 1116 cp += off + 2 * sizeof(uint16_t); 1117 totlen -= 2 * sizeof(uint16_t); 1118 } 1119 MGETHDR(m, M_DONTWAIT, MT_DATA); 1120 if (m == 0) 1121 return (NULL); 1122 m->m_pkthdr.rcvif = ifp; 1123 m->m_pkthdr.len = totlen; 1124 m->m_len = MHLEN; 1125 1126 while (totlen > 0) { 1127 if (top) { 1128 MGET(m, M_DONTWAIT, MT_DATA); 1129 if (m == 0) { 1130 m_freem(top); 1131 return (NULL); 1132 } 1133 m->m_len = MLEN; 1134 } 1135 len = min(totlen, epkt - cp); 1136 if (len >= MINCLSIZE) { 1137 MCLGET(m, M_DONTWAIT); 1138 if ((m->m_flags & M_EXT) == 0) { 1139 m_free(m); 1140 m_freem(top); 1141 return (NULL); 1142 } 1143 m->m_len = len = min(len, MCLBYTES); 1144 } else { 1145 /* 1146 * Place initial small packet/header at end of mbuf. 1147 */ 1148 if (len < m->m_len) { 1149 if (top == 0 && len + max_linkhdr <= m->m_len) 1150 m->m_data += max_linkhdr; 1151 m->m_len = len; 1152 } else 1153 len = m->m_len; 1154 } 1155 if (copy) 1156 copy(cp, mtod(m, void *), (size_t)len); 1157 else 1158 memcpy(mtod(m, void *), cp, (size_t)len); 1159 cp += len; 1160 *mp = m; 1161 mp = &m->m_next; 1162 totlen -= len; 1163 if (cp == epkt) 1164 cp = buf; 1165 } 1166 return (top); 1167 } 1168 1169 /* 1170 * Copy data from a buffer back into the indicated mbuf chain, 1171 * starting "off" bytes from the beginning, extending the mbuf 1172 * chain if necessary. 1173 */ 1174 void 1175 m_copyback(struct mbuf *m0, int off, int len, const void *cp) 1176 { 1177 #if defined(DEBUG) 1178 struct mbuf *origm = m0; 1179 int error; 1180 #endif /* defined(DEBUG) */ 1181 1182 if (m0 == NULL) 1183 return; 1184 1185 #if defined(DEBUG) 1186 error = 1187 #endif /* defined(DEBUG) */ 1188 m_copyback0(&m0, off, len, cp, 1189 M_COPYBACK0_COPYBACK|M_COPYBACK0_EXTEND, M_DONTWAIT); 1190 1191 #if defined(DEBUG) 1192 if (error != 0 || (m0 != NULL && origm != m0)) 1193 panic("m_copyback"); 1194 #endif /* defined(DEBUG) */ 1195 } 1196 1197 struct mbuf * 1198 m_copyback_cow(struct mbuf *m0, int off, int len, const void *cp, int how) 1199 { 1200 int error; 1201 1202 /* don't support chain expansion */ 1203 KDASSERT(off + len <= m_length(m0)); 1204 1205 error = m_copyback0(&m0, off, len, cp, 1206 M_COPYBACK0_COPYBACK|M_COPYBACK0_COW, how); 1207 if (error) { 1208 /* 1209 * no way to recover from partial success. 1210 * just free the chain. 1211 */ 1212 m_freem(m0); 1213 return NULL; 1214 } 1215 return m0; 1216 } 1217 1218 /* 1219 * m_makewritable: ensure the specified range writable. 1220 */ 1221 int 1222 m_makewritable(struct mbuf **mp, int off, int len, int how) 1223 { 1224 int error; 1225 #if defined(DEBUG) 1226 struct mbuf *n; 1227 int origlen, reslen; 1228 1229 origlen = m_length(*mp); 1230 #endif /* defined(DEBUG) */ 1231 1232 #if 0 /* M_COPYALL is large enough */ 1233 if (len == M_COPYALL) 1234 len = m_length(*mp) - off; /* XXX */ 1235 #endif 1236 1237 error = m_copyback0(mp, off, len, NULL, 1238 M_COPYBACK0_PRESERVE|M_COPYBACK0_COW, how); 1239 1240 #if defined(DEBUG) 1241 reslen = 0; 1242 for (n = *mp; n; n = n->m_next) 1243 reslen += n->m_len; 1244 if (origlen != reslen) 1245 panic("m_makewritable: length changed"); 1246 if (((*mp)->m_flags & M_PKTHDR) != 0 && reslen != (*mp)->m_pkthdr.len) 1247 panic("m_makewritable: inconsist"); 1248 #endif /* defined(DEBUG) */ 1249 1250 return error; 1251 } 1252 1253 /* 1254 * Copy the mbuf chain to a new mbuf chain that is as short as possible. 1255 * Return the new mbuf chain on success, NULL on failure. On success, 1256 * free the old mbuf chain. 1257 */ 1258 struct mbuf * 1259 m_defrag(struct mbuf *mold, int flags) 1260 { 1261 struct mbuf *m0, *mn, *n; 1262 size_t sz = mold->m_pkthdr.len; 1263 1264 #ifdef DIAGNOSTIC 1265 if ((mold->m_flags & M_PKTHDR) == 0) 1266 panic("m_defrag: not a mbuf chain header"); 1267 #endif 1268 1269 MGETHDR(m0, flags, MT_DATA); 1270 if (m0 == NULL) 1271 return NULL; 1272 M_COPY_PKTHDR(m0, mold); 1273 mn = m0; 1274 1275 do { 1276 if (sz > MHLEN) { 1277 MCLGET(mn, M_DONTWAIT); 1278 if ((mn->m_flags & M_EXT) == 0) { 1279 m_freem(m0); 1280 return NULL; 1281 } 1282 } 1283 1284 mn->m_len = MIN(sz, MCLBYTES); 1285 1286 m_copydata(mold, mold->m_pkthdr.len - sz, mn->m_len, 1287 mtod(mn, void *)); 1288 1289 sz -= mn->m_len; 1290 1291 if (sz > 0) { 1292 /* need more mbufs */ 1293 MGET(n, M_NOWAIT, MT_DATA); 1294 if (n == NULL) { 1295 m_freem(m0); 1296 return NULL; 1297 } 1298 1299 mn->m_next = n; 1300 mn = n; 1301 } 1302 } while (sz > 0); 1303 1304 m_freem(mold); 1305 1306 return m0; 1307 } 1308 1309 int 1310 m_copyback0(struct mbuf **mp0, int off, int len, const void *vp, int flags, 1311 int how) 1312 { 1313 int mlen; 1314 struct mbuf *m, *n; 1315 struct mbuf **mp; 1316 int totlen = 0; 1317 const char *cp = vp; 1318 1319 KASSERT(mp0 != NULL); 1320 KASSERT(*mp0 != NULL); 1321 KASSERT((flags & M_COPYBACK0_PRESERVE) == 0 || cp == NULL); 1322 KASSERT((flags & M_COPYBACK0_COPYBACK) == 0 || cp != NULL); 1323 1324 /* 1325 * we don't bother to update "totlen" in the case of M_COPYBACK0_COW, 1326 * assuming that M_COPYBACK0_EXTEND and M_COPYBACK0_COW are exclusive. 1327 */ 1328 1329 KASSERT((~flags & (M_COPYBACK0_EXTEND|M_COPYBACK0_COW)) != 0); 1330 1331 mp = mp0; 1332 m = *mp; 1333 while (off > (mlen = m->m_len)) { 1334 off -= mlen; 1335 totlen += mlen; 1336 if (m->m_next == NULL) { 1337 int tspace; 1338 extend: 1339 if ((flags & M_COPYBACK0_EXTEND) == 0) 1340 goto out; 1341 1342 /* 1343 * try to make some space at the end of "m". 1344 */ 1345 1346 mlen = m->m_len; 1347 if (off + len >= MINCLSIZE && 1348 (m->m_flags & M_EXT) == 0 && m->m_len == 0) { 1349 MCLGET(m, how); 1350 } 1351 tspace = M_TRAILINGSPACE(m); 1352 if (tspace > 0) { 1353 tspace = min(tspace, off + len); 1354 KASSERT(tspace > 0); 1355 memset(mtod(m, char *) + m->m_len, 0, 1356 min(off, tspace)); 1357 m->m_len += tspace; 1358 off += mlen; 1359 totlen -= mlen; 1360 continue; 1361 } 1362 1363 /* 1364 * need to allocate an mbuf. 1365 */ 1366 1367 if (off + len >= MINCLSIZE) { 1368 n = m_getcl(how, m->m_type, 0); 1369 } else { 1370 n = m_get(how, m->m_type); 1371 } 1372 if (n == NULL) { 1373 goto out; 1374 } 1375 n->m_len = 0; 1376 n->m_len = min(M_TRAILINGSPACE(n), off + len); 1377 memset(mtod(n, char *), 0, min(n->m_len, off)); 1378 m->m_next = n; 1379 } 1380 mp = &m->m_next; 1381 m = m->m_next; 1382 } 1383 while (len > 0) { 1384 mlen = m->m_len - off; 1385 if (mlen != 0 && M_READONLY(m)) { 1386 char *datap; 1387 int eatlen; 1388 1389 /* 1390 * this mbuf is read-only. 1391 * allocate a new writable mbuf and try again. 1392 */ 1393 1394 #if defined(DIAGNOSTIC) 1395 if ((flags & M_COPYBACK0_COW) == 0) 1396 panic("m_copyback0: read-only"); 1397 #endif /* defined(DIAGNOSTIC) */ 1398 1399 /* 1400 * if we're going to write into the middle of 1401 * a mbuf, split it first. 1402 */ 1403 if (off > 0) { 1404 n = m_split0(m, off, how, 0); 1405 if (n == NULL) 1406 goto enobufs; 1407 m->m_next = n; 1408 mp = &m->m_next; 1409 m = n; 1410 off = 0; 1411 continue; 1412 } 1413 1414 /* 1415 * XXX TODO coalesce into the trailingspace of 1416 * the previous mbuf when possible. 1417 */ 1418 1419 /* 1420 * allocate a new mbuf. copy packet header if needed. 1421 */ 1422 MGET(n, how, m->m_type); 1423 if (n == NULL) 1424 goto enobufs; 1425 MCLAIM(n, m->m_owner); 1426 if (off == 0 && (m->m_flags & M_PKTHDR) != 0) { 1427 M_MOVE_PKTHDR(n, m); 1428 n->m_len = MHLEN; 1429 } else { 1430 if (len >= MINCLSIZE) 1431 MCLGET(n, M_DONTWAIT); 1432 n->m_len = 1433 (n->m_flags & M_EXT) ? MCLBYTES : MLEN; 1434 } 1435 if (n->m_len > len) 1436 n->m_len = len; 1437 1438 /* 1439 * free the region which has been overwritten. 1440 * copying data from old mbufs if requested. 1441 */ 1442 if (flags & M_COPYBACK0_PRESERVE) 1443 datap = mtod(n, char *); 1444 else 1445 datap = NULL; 1446 eatlen = n->m_len; 1447 while (m != NULL && M_READONLY(m) && 1448 n->m_type == m->m_type && eatlen > 0) { 1449 mlen = min(eatlen, m->m_len); 1450 if (datap) { 1451 m_copydata(m, 0, mlen, datap); 1452 datap += mlen; 1453 } 1454 m->m_data += mlen; 1455 m->m_len -= mlen; 1456 eatlen -= mlen; 1457 if (m->m_len == 0) 1458 *mp = m = m_free(m); 1459 } 1460 if (eatlen > 0) 1461 n->m_len -= eatlen; 1462 n->m_next = m; 1463 *mp = m = n; 1464 continue; 1465 } 1466 mlen = min(mlen, len); 1467 if (flags & M_COPYBACK0_COPYBACK) { 1468 memcpy(mtod(m, char *) + off, cp, (unsigned)mlen); 1469 cp += mlen; 1470 } 1471 len -= mlen; 1472 mlen += off; 1473 off = 0; 1474 totlen += mlen; 1475 if (len == 0) 1476 break; 1477 if (m->m_next == NULL) { 1478 goto extend; 1479 } 1480 mp = &m->m_next; 1481 m = m->m_next; 1482 } 1483 out: if (((m = *mp0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen)) { 1484 KASSERT((flags & M_COPYBACK0_EXTEND) != 0); 1485 m->m_pkthdr.len = totlen; 1486 } 1487 1488 return 0; 1489 1490 enobufs: 1491 return ENOBUFS; 1492 } 1493 1494 void 1495 m_move_pkthdr(struct mbuf *to, struct mbuf *from) 1496 { 1497 1498 KASSERT((to->m_flags & M_EXT) == 0); 1499 KASSERT((to->m_flags & M_PKTHDR) == 0 || m_tag_first(to) == NULL); 1500 KASSERT((from->m_flags & M_PKTHDR) != 0); 1501 1502 to->m_pkthdr = from->m_pkthdr; 1503 to->m_flags = from->m_flags & M_COPYFLAGS; 1504 to->m_data = to->m_pktdat; 1505 1506 from->m_flags &= ~M_PKTHDR; 1507 } 1508 1509 /* 1510 * Apply function f to the data in an mbuf chain starting "off" bytes from the 1511 * beginning, continuing for "len" bytes. 1512 */ 1513 int 1514 m_apply(struct mbuf *m, int off, int len, 1515 int (*f)(void *, void *, unsigned int), void *arg) 1516 { 1517 unsigned int count; 1518 int rval; 1519 1520 KASSERT(len >= 0); 1521 KASSERT(off >= 0); 1522 1523 while (off > 0) { 1524 KASSERT(m != NULL); 1525 if (off < m->m_len) 1526 break; 1527 off -= m->m_len; 1528 m = m->m_next; 1529 } 1530 while (len > 0) { 1531 KASSERT(m != NULL); 1532 count = min(m->m_len - off, len); 1533 1534 rval = (*f)(arg, mtod(m, char *) + off, count); 1535 if (rval) 1536 return (rval); 1537 1538 len -= count; 1539 off = 0; 1540 m = m->m_next; 1541 } 1542 1543 return (0); 1544 } 1545 1546 /* 1547 * Return a pointer to mbuf/offset of location in mbuf chain. 1548 */ 1549 struct mbuf * 1550 m_getptr(struct mbuf *m, int loc, int *off) 1551 { 1552 1553 while (loc >= 0) { 1554 /* Normal end of search */ 1555 if (m->m_len > loc) { 1556 *off = loc; 1557 return (m); 1558 } else { 1559 loc -= m->m_len; 1560 1561 if (m->m_next == NULL) { 1562 if (loc == 0) { 1563 /* Point at the end of valid data */ 1564 *off = m->m_len; 1565 return (m); 1566 } else 1567 return (NULL); 1568 } else 1569 m = m->m_next; 1570 } 1571 } 1572 1573 return (NULL); 1574 } 1575 1576 /* 1577 * m_ext_free: release a reference to the mbuf external storage. 1578 * 1579 * => free the mbuf m itsself as well. 1580 */ 1581 1582 void 1583 m_ext_free(struct mbuf *m) 1584 { 1585 bool embedded = MEXT_ISEMBEDDED(m); 1586 bool dofree = true; 1587 u_int refcnt; 1588 1589 KASSERT((m->m_flags & M_EXT) != 0); 1590 KASSERT(MEXT_ISEMBEDDED(m->m_ext_ref)); 1591 KASSERT((m->m_ext_ref->m_flags & M_EXT) != 0); 1592 KASSERT((m->m_flags & M_EXT_CLUSTER) == 1593 (m->m_ext_ref->m_flags & M_EXT_CLUSTER)); 1594 1595 if (__predict_true(m->m_ext.ext_refcnt == 1)) { 1596 refcnt = m->m_ext.ext_refcnt = 0; 1597 } else { 1598 refcnt = atomic_dec_uint_nv(&m->m_ext.ext_refcnt); 1599 } 1600 if (refcnt > 0) { 1601 if (embedded) { 1602 /* 1603 * other mbuf's m_ext_ref still points to us. 1604 */ 1605 dofree = false; 1606 } else { 1607 m->m_ext_ref = m; 1608 } 1609 } else { 1610 /* 1611 * dropping the last reference 1612 */ 1613 if (!embedded) { 1614 m->m_ext.ext_refcnt++; /* XXX */ 1615 m_ext_free(m->m_ext_ref); 1616 m->m_ext_ref = m; 1617 } else if ((m->m_flags & M_EXT_CLUSTER) != 0) { 1618 pool_cache_put_paddr((struct pool_cache *) 1619 m->m_ext.ext_arg, 1620 m->m_ext.ext_buf, m->m_ext.ext_paddr); 1621 } else if (m->m_ext.ext_free) { 1622 (*m->m_ext.ext_free)(m, 1623 m->m_ext.ext_buf, m->m_ext.ext_size, 1624 m->m_ext.ext_arg); 1625 /* 1626 * 'm' is already freed by the ext_free callback. 1627 */ 1628 dofree = false; 1629 } else { 1630 free(m->m_ext.ext_buf, m->m_ext.ext_type); 1631 } 1632 } 1633 if (dofree) { 1634 pool_cache_put(mb_cache, m); 1635 } 1636 } 1637 1638 #if defined(DDB) 1639 void 1640 m_print(const struct mbuf *m, const char *modif, void (*pr)(const char *, ...)) 1641 { 1642 char ch; 1643 bool opt_c = false; 1644 char buf[512]; 1645 1646 while ((ch = *(modif++)) != '\0') { 1647 switch (ch) { 1648 case 'c': 1649 opt_c = true; 1650 break; 1651 } 1652 } 1653 1654 nextchain: 1655 (*pr)("MBUF %p\n", m); 1656 snprintb(buf, sizeof(buf), M_FLAGS_BITS, (u_int)m->m_flags); 1657 (*pr)(" data=%p, len=%d, type=%d, flags=%s\n", 1658 m->m_data, m->m_len, m->m_type, buf); 1659 (*pr)(" owner=%p, next=%p, nextpkt=%p\n", m->m_owner, m->m_next, 1660 m->m_nextpkt); 1661 (*pr)(" leadingspace=%u, trailingspace=%u, readonly=%u\n", 1662 (int)M_LEADINGSPACE(m), (int)M_TRAILINGSPACE(m), 1663 (int)M_READONLY(m)); 1664 if ((m->m_flags & M_PKTHDR) != 0) { 1665 snprintb(buf, sizeof(buf), M_CSUM_BITS, m->m_pkthdr.csum_flags); 1666 (*pr)(" pktlen=%d, rcvif=%p, csum_flags=0x%s, csum_data=0x%" 1667 PRIx32 ", segsz=%u\n", 1668 m->m_pkthdr.len, m->m_pkthdr.rcvif, 1669 buf, m->m_pkthdr.csum_data, m->m_pkthdr.segsz); 1670 } 1671 if ((m->m_flags & M_EXT)) { 1672 (*pr)(" ext_refcnt=%u, ext_buf=%p, ext_size=%zd, " 1673 "ext_free=%p, ext_arg=%p\n", 1674 m->m_ext.ext_refcnt, 1675 m->m_ext.ext_buf, m->m_ext.ext_size, 1676 m->m_ext.ext_free, m->m_ext.ext_arg); 1677 } 1678 if ((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 0) { 1679 vaddr_t sva = (vaddr_t)m->m_ext.ext_buf; 1680 vaddr_t eva = sva + m->m_ext.ext_size; 1681 int n = (round_page(eva) - trunc_page(sva)) >> PAGE_SHIFT; 1682 int i; 1683 1684 (*pr)(" pages:"); 1685 for (i = 0; i < n; i ++) { 1686 (*pr)(" %p", m->m_ext.ext_pgs[i]); 1687 } 1688 (*pr)("\n"); 1689 } 1690 1691 if (opt_c) { 1692 m = m->m_next; 1693 if (m != NULL) { 1694 goto nextchain; 1695 } 1696 } 1697 } 1698 #endif /* defined(DDB) */ 1699 1700 void 1701 mbstat_type_add(int type, int diff) 1702 { 1703 struct mbstat_cpu *mb; 1704 int s; 1705 1706 s = splvm(); 1707 mb = percpu_getref(mbstat_percpu); 1708 mb->m_mtypes[type] += diff; 1709 percpu_putref(mbstat_percpu); 1710 splx(s); 1711 } 1712 1713 #if defined(MBUFTRACE) 1714 void 1715 mowner_attach(struct mowner *mo) 1716 { 1717 1718 KASSERT(mo->mo_counters == NULL); 1719 mo->mo_counters = percpu_alloc(sizeof(struct mowner_counter)); 1720 1721 /* XXX lock */ 1722 LIST_INSERT_HEAD(&mowners, mo, mo_link); 1723 } 1724 1725 void 1726 mowner_detach(struct mowner *mo) 1727 { 1728 1729 KASSERT(mo->mo_counters != NULL); 1730 1731 /* XXX lock */ 1732 LIST_REMOVE(mo, mo_link); 1733 1734 percpu_free(mo->mo_counters, sizeof(struct mowner_counter)); 1735 mo->mo_counters = NULL; 1736 } 1737 1738 void 1739 mowner_init(struct mbuf *m, int type) 1740 { 1741 struct mowner_counter *mc; 1742 struct mowner *mo; 1743 int s; 1744 1745 m->m_owner = mo = &unknown_mowners[type]; 1746 s = splvm(); 1747 mc = percpu_getref(mo->mo_counters); 1748 mc->mc_counter[MOWNER_COUNTER_CLAIMS]++; 1749 percpu_putref(mo->mo_counters); 1750 splx(s); 1751 } 1752 1753 void 1754 mowner_ref(struct mbuf *m, int flags) 1755 { 1756 struct mowner *mo = m->m_owner; 1757 struct mowner_counter *mc; 1758 int s; 1759 1760 s = splvm(); 1761 mc = percpu_getref(mo->mo_counters); 1762 if ((flags & M_EXT) != 0) 1763 mc->mc_counter[MOWNER_COUNTER_EXT_CLAIMS]++; 1764 if ((flags & M_CLUSTER) != 0) 1765 mc->mc_counter[MOWNER_COUNTER_CLUSTER_CLAIMS]++; 1766 percpu_putref(mo->mo_counters); 1767 splx(s); 1768 } 1769 1770 void 1771 mowner_revoke(struct mbuf *m, bool all, int flags) 1772 { 1773 struct mowner *mo = m->m_owner; 1774 struct mowner_counter *mc; 1775 int s; 1776 1777 s = splvm(); 1778 mc = percpu_getref(mo->mo_counters); 1779 if ((flags & M_EXT) != 0) 1780 mc->mc_counter[MOWNER_COUNTER_EXT_RELEASES]++; 1781 if ((flags & M_CLUSTER) != 0) 1782 mc->mc_counter[MOWNER_COUNTER_CLUSTER_RELEASES]++; 1783 if (all) 1784 mc->mc_counter[MOWNER_COUNTER_RELEASES]++; 1785 percpu_putref(mo->mo_counters); 1786 splx(s); 1787 if (all) 1788 m->m_owner = &revoked_mowner; 1789 } 1790 1791 static void 1792 mowner_claim(struct mbuf *m, struct mowner *mo) 1793 { 1794 struct mowner_counter *mc; 1795 int flags = m->m_flags; 1796 int s; 1797 1798 s = splvm(); 1799 mc = percpu_getref(mo->mo_counters); 1800 mc->mc_counter[MOWNER_COUNTER_CLAIMS]++; 1801 if ((flags & M_EXT) != 0) 1802 mc->mc_counter[MOWNER_COUNTER_EXT_CLAIMS]++; 1803 if ((flags & M_CLUSTER) != 0) 1804 mc->mc_counter[MOWNER_COUNTER_CLUSTER_CLAIMS]++; 1805 percpu_putref(mo->mo_counters); 1806 splx(s); 1807 m->m_owner = mo; 1808 } 1809 1810 void 1811 m_claim(struct mbuf *m, struct mowner *mo) 1812 { 1813 1814 if (m->m_owner == mo || mo == NULL) 1815 return; 1816 1817 mowner_revoke(m, true, m->m_flags); 1818 mowner_claim(m, mo); 1819 } 1820 #endif /* defined(MBUFTRACE) */ 1821