1 /* 2 * (MPSAFE) 3 * 4 * Copyright (c) 2004 Jeffrey M. Hsu. All rights reserved. 5 * Copyright (c) 2004 The DragonFly Project. All rights reserved. 6 * 7 * This code is derived from software contributed to The DragonFly Project 8 * by Jeffrey M. Hsu. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of The DragonFly Project nor the names of its 19 * contributors may be used to endorse or promote products derived 20 * from this software without specific, prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 23 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 25 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 26 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 27 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 28 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 29 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 30 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 31 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 32 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 33 * SUCH DAMAGE. 34 */ 35 36 /* 37 * Copyright (c) 1982, 1986, 1988, 1991, 1993 38 * The Regents of the University of California. All rights reserved. 39 * 40 * Redistribution and use in source and binary forms, with or without 41 * modification, are permitted provided that the following conditions 42 * are met: 43 * 1. Redistributions of source code must retain the above copyright 44 * notice, this list of conditions and the following disclaimer. 45 * 2. Redistributions in binary form must reproduce the above copyright 46 * notice, this list of conditions and the following disclaimer in the 47 * documentation and/or other materials provided with the distribution. 48 * 3. Neither the name of the University nor the names of its contributors 49 * may be used to endorse or promote products derived from this software 50 * without specific prior written permission. 51 * 52 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 53 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 54 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 55 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 56 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 57 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 58 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 59 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 60 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 61 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 62 * SUCH DAMAGE. 63 * 64 * @(#)uipc_mbuf.c 8.2 (Berkeley) 1/4/94 65 * $FreeBSD: src/sys/kern/uipc_mbuf.c,v 1.51.2.24 2003/04/15 06:59:29 silby Exp $ 66 */ 67 68 #include "opt_param.h" 69 #include "opt_mbuf_stress_test.h" 70 #include <sys/param.h> 71 #include <sys/systm.h> 72 #include <sys/file.h> 73 #include <sys/malloc.h> 74 #include <sys/mbuf.h> 75 #include <sys/kernel.h> 76 #include <sys/sysctl.h> 77 #include <sys/domain.h> 78 #include <sys/objcache.h> 79 #include <sys/tree.h> 80 #include <sys/protosw.h> 81 #include <sys/uio.h> 82 #include <sys/thread.h> 83 #include <sys/globaldata.h> 84 85 #include <sys/thread2.h> 86 #include <sys/spinlock2.h> 87 88 #include <machine/atomic.h> 89 #include <machine/limits.h> 90 91 #include <vm/vm.h> 92 #include <vm/vm_kern.h> 93 #include <vm/vm_extern.h> 94 95 #ifdef INVARIANTS 96 #include <machine/cpu.h> 97 #endif 98 99 /* 100 * mbuf cluster meta-data 101 */ 102 struct mbcluster { 103 int32_t mcl_refs; 104 void *mcl_data; 105 }; 106 107 /* 108 * mbuf tracking for debugging purposes 109 */ 110 #ifdef MBUF_DEBUG 111 112 static MALLOC_DEFINE(M_MTRACK, "mtrack", "mtrack"); 113 114 struct mbctrack; 115 RB_HEAD(mbuf_rb_tree, mbtrack); 116 RB_PROTOTYPE2(mbuf_rb_tree, mbtrack, rb_node, mbtrack_cmp, struct mbuf *); 117 118 struct mbtrack { 119 RB_ENTRY(mbtrack) rb_node; 120 int trackid; 121 struct mbuf *m; 122 }; 123 124 static int 125 mbtrack_cmp(struct mbtrack *mb1, struct mbtrack *mb2) 126 { 127 if (mb1->m < mb2->m) 128 return(-1); 129 if (mb1->m > mb2->m) 130 return(1); 131 return(0); 132 } 133 134 RB_GENERATE2(mbuf_rb_tree, mbtrack, rb_node, mbtrack_cmp, struct mbuf *, m); 135 136 struct mbuf_rb_tree mbuf_track_root; 137 static struct spinlock mbuf_track_spin = SPINLOCK_INITIALIZER(mbuf_track_spin, "mbuf_track_spin"); 138 139 static void 140 mbuftrack(struct mbuf *m) 141 { 142 struct mbtrack *mbt; 143 144 mbt = kmalloc(sizeof(*mbt), M_MTRACK, M_INTWAIT|M_ZERO); 145 spin_lock(&mbuf_track_spin); 146 mbt->m = m; 147 if (mbuf_rb_tree_RB_INSERT(&mbuf_track_root, mbt)) { 148 spin_unlock(&mbuf_track_spin); 149 panic("mbuftrack: mbuf %p already being tracked", m); 150 } 151 spin_unlock(&mbuf_track_spin); 152 } 153 154 static void 155 mbufuntrack(struct mbuf *m) 156 { 157 struct mbtrack *mbt; 158 159 spin_lock(&mbuf_track_spin); 160 mbt = mbuf_rb_tree_RB_LOOKUP(&mbuf_track_root, m); 161 if (mbt == NULL) { 162 spin_unlock(&mbuf_track_spin); 163 panic("mbufuntrack: mbuf %p was not tracked", m); 164 } else { 165 mbuf_rb_tree_RB_REMOVE(&mbuf_track_root, mbt); 166 spin_unlock(&mbuf_track_spin); 167 kfree(mbt, M_MTRACK); 168 } 169 } 170 171 void 172 mbuftrackid(struct mbuf *m, int trackid) 173 { 174 struct mbtrack *mbt; 175 struct mbuf *n; 176 177 spin_lock(&mbuf_track_spin); 178 while (m) { 179 n = m->m_nextpkt; 180 while (m) { 181 mbt = mbuf_rb_tree_RB_LOOKUP(&mbuf_track_root, m); 182 if (mbt == NULL) { 183 spin_unlock(&mbuf_track_spin); 184 panic("mbuftrackid: mbuf %p not tracked", m); 185 } 186 mbt->trackid = trackid; 187 m = m->m_next; 188 } 189 m = n; 190 } 191 spin_unlock(&mbuf_track_spin); 192 } 193 194 static int 195 mbuftrack_callback(struct mbtrack *mbt, void *arg) 196 { 197 struct sysctl_req *req = arg; 198 char buf[64]; 199 int error; 200 201 ksnprintf(buf, sizeof(buf), "mbuf %p track %d\n", mbt->m, mbt->trackid); 202 203 spin_unlock(&mbuf_track_spin); 204 error = SYSCTL_OUT(req, buf, strlen(buf)); 205 spin_lock(&mbuf_track_spin); 206 if (error) 207 return(-error); 208 return(0); 209 } 210 211 static int 212 mbuftrack_show(SYSCTL_HANDLER_ARGS) 213 { 214 int error; 215 216 spin_lock(&mbuf_track_spin); 217 error = mbuf_rb_tree_RB_SCAN(&mbuf_track_root, NULL, 218 mbuftrack_callback, req); 219 spin_unlock(&mbuf_track_spin); 220 return (-error); 221 } 222 SYSCTL_PROC(_kern_ipc, OID_AUTO, showmbufs, CTLFLAG_RD|CTLTYPE_STRING, 223 0, 0, mbuftrack_show, "A", "Show all in-use mbufs"); 224 225 #else 226 227 #define mbuftrack(m) 228 #define mbufuntrack(m) 229 230 #endif 231 232 static void mbinit(void *); 233 SYSINIT(mbuf, SI_BOOT2_MACHDEP, SI_ORDER_FIRST, mbinit, NULL); 234 235 struct mbtypes_stat { 236 u_long stats[MT_NTYPES]; 237 } __cachealign; 238 239 static struct mbtypes_stat mbtypes[SMP_MAXCPU]; 240 241 static struct mbstat mbstat[SMP_MAXCPU] __cachealign; 242 int max_linkhdr; 243 int max_protohdr; 244 int max_hdr; 245 int max_datalen; 246 int m_defragpackets; 247 int m_defragbytes; 248 int m_defraguseless; 249 int m_defragfailure; 250 #ifdef MBUF_STRESS_TEST 251 int m_defragrandomfailures; 252 #endif 253 254 struct objcache *mbuf_cache, *mbufphdr_cache; 255 struct objcache *mclmeta_cache, *mjclmeta_cache; 256 struct objcache *mbufcluster_cache, *mbufphdrcluster_cache; 257 struct objcache *mbufjcluster_cache, *mbufphdrjcluster_cache; 258 259 struct lock mbupdate_lk = LOCK_INITIALIZER("mbupdate", 0, LK_CANRECURSE); 260 261 int nmbclusters; 262 static int nmbjclusters; 263 int nmbufs; 264 265 static int mjclph_cachefrac; 266 static int mjcl_cachefrac; 267 static int mclph_cachefrac; 268 static int mcl_cachefrac; 269 270 SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RW, 271 &max_linkhdr, 0, "Max size of a link-level header"); 272 SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RW, 273 &max_protohdr, 0, "Max size of a protocol header"); 274 SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RW, &max_hdr, 0, 275 "Max size of link+protocol headers"); 276 SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RW, 277 &max_datalen, 0, "Max data payload size without headers"); 278 SYSCTL_INT(_kern_ipc, OID_AUTO, mbuf_wait, CTLFLAG_RW, 279 &mbuf_wait, 0, "Time in ticks to sleep after failed mbuf allocations"); 280 static int do_mbstat(SYSCTL_HANDLER_ARGS); 281 282 SYSCTL_PROC(_kern_ipc, KIPC_MBSTAT, mbstat, CTLTYPE_STRUCT|CTLFLAG_RD, 283 0, 0, do_mbstat, "S,mbstat", "mbuf usage statistics"); 284 285 static int do_mbtypes(SYSCTL_HANDLER_ARGS); 286 287 SYSCTL_PROC(_kern_ipc, OID_AUTO, mbtypes, CTLTYPE_ULONG|CTLFLAG_RD, 288 0, 0, do_mbtypes, "LU", ""); 289 290 static int 291 do_mbstat(SYSCTL_HANDLER_ARGS) 292 { 293 struct mbstat mbstat_total; 294 struct mbstat *mbstat_totalp; 295 int i; 296 297 bzero(&mbstat_total, sizeof(mbstat_total)); 298 mbstat_totalp = &mbstat_total; 299 300 for (i = 0; i < ncpus; i++) { 301 mbstat_total.m_mbufs += mbstat[i].m_mbufs; 302 mbstat_total.m_clusters += mbstat[i].m_clusters; 303 mbstat_total.m_jclusters += mbstat[i].m_jclusters; 304 mbstat_total.m_clfree += mbstat[i].m_clfree; 305 mbstat_total.m_drops += mbstat[i].m_drops; 306 mbstat_total.m_wait += mbstat[i].m_wait; 307 mbstat_total.m_drain += mbstat[i].m_drain; 308 mbstat_total.m_mcfail += mbstat[i].m_mcfail; 309 mbstat_total.m_mpfail += mbstat[i].m_mpfail; 310 311 } 312 /* 313 * The following fields are not cumulative fields so just 314 * get their values once. 315 */ 316 mbstat_total.m_msize = mbstat[0].m_msize; 317 mbstat_total.m_mclbytes = mbstat[0].m_mclbytes; 318 mbstat_total.m_minclsize = mbstat[0].m_minclsize; 319 mbstat_total.m_mlen = mbstat[0].m_mlen; 320 mbstat_total.m_mhlen = mbstat[0].m_mhlen; 321 322 return(sysctl_handle_opaque(oidp, mbstat_totalp, sizeof(mbstat_total), req)); 323 } 324 325 static int 326 do_mbtypes(SYSCTL_HANDLER_ARGS) 327 { 328 u_long totals[MT_NTYPES]; 329 int i, j; 330 331 for (i = 0; i < MT_NTYPES; i++) 332 totals[i] = 0; 333 334 for (i = 0; i < ncpus; i++) { 335 for (j = 0; j < MT_NTYPES; j++) 336 totals[j] += mbtypes[i].stats[j]; 337 } 338 339 return(sysctl_handle_opaque(oidp, totals, sizeof(totals), req)); 340 } 341 342 /* 343 * The variables may be set as boot-time tunables or live. Setting these 344 * values too low can deadlock your network. Network interfaces may also 345 * adjust nmbclusters and/or nmbjclusters to account for preloading the 346 * hardware rings. 347 */ 348 static int sysctl_nmbclusters(SYSCTL_HANDLER_ARGS); 349 static int sysctl_nmbjclusters(SYSCTL_HANDLER_ARGS); 350 static int sysctl_nmbufs(SYSCTL_HANDLER_ARGS); 351 SYSCTL_PROC(_kern_ipc, KIPC_NMBCLUSTERS, nmbclusters, CTLTYPE_INT | CTLFLAG_RW, 352 0, 0, sysctl_nmbclusters, "I", 353 "Maximum number of mbuf clusters available"); 354 SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjclusters, CTLTYPE_INT | CTLFLAG_RW, 355 0, 0, sysctl_nmbjclusters, "I", 356 "Maximum number of mbuf jclusters available"); 357 SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbufs, CTLTYPE_INT | CTLFLAG_RW, 358 0, 0, sysctl_nmbufs, "I", 359 "Maximum number of mbufs available"); 360 361 SYSCTL_INT(_kern_ipc, OID_AUTO, mjclph_cachefrac, CTLFLAG_RD, 362 &mjclph_cachefrac, 0, 363 "Fraction of cacheable mbuf jclusters w/ pkthdr"); 364 SYSCTL_INT(_kern_ipc, OID_AUTO, mjcl_cachefrac, CTLFLAG_RD, 365 &mjcl_cachefrac, 0, 366 "Fraction of cacheable mbuf jclusters"); 367 SYSCTL_INT(_kern_ipc, OID_AUTO, mclph_cachefrac, CTLFLAG_RD, 368 &mclph_cachefrac, 0, 369 "Fraction of cacheable mbuf clusters w/ pkthdr"); 370 SYSCTL_INT(_kern_ipc, OID_AUTO, mcl_cachefrac, CTLFLAG_RD, 371 &mcl_cachefrac, 0, "Fraction of cacheable mbuf clusters"); 372 373 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD, 374 &m_defragpackets, 0, "Number of defragment packets"); 375 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD, 376 &m_defragbytes, 0, "Number of defragment bytes"); 377 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD, 378 &m_defraguseless, 0, "Number of useless defragment mbuf chain operations"); 379 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD, 380 &m_defragfailure, 0, "Number of failed defragment mbuf chain operations"); 381 #ifdef MBUF_STRESS_TEST 382 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW, 383 &m_defragrandomfailures, 0, ""); 384 #endif 385 386 static MALLOC_DEFINE(M_MBUF, "mbuf", "mbuf"); 387 static MALLOC_DEFINE(M_MBUFCL, "mbufcl", "mbufcl"); 388 static MALLOC_DEFINE(M_MCLMETA, "mclmeta", "mclmeta"); 389 390 static void m_reclaim (void); 391 static void m_mclref(void *arg); 392 static void m_mclfree(void *arg); 393 static void m_mjclfree(void *arg); 394 395 static void mbupdatelimits(void); 396 397 /* 398 * NOTE: Default NMBUFS must take into account a possible DOS attack 399 * using fd passing on unix domain sockets. 400 */ 401 #ifndef NMBCLUSTERS 402 #define NMBCLUSTERS (512 + maxusers * 16) 403 #endif 404 #ifndef MJCLPH_CACHEFRAC 405 #define MJCLPH_CACHEFRAC 16 406 #endif 407 #ifndef MJCL_CACHEFRAC 408 #define MJCL_CACHEFRAC 4 409 #endif 410 #ifndef MCLPH_CACHEFRAC 411 #define MCLPH_CACHEFRAC 16 412 #endif 413 #ifndef MCL_CACHEFRAC 414 #define MCL_CACHEFRAC 4 415 #endif 416 #ifndef NMBJCLUSTERS 417 #define NMBJCLUSTERS (NMBCLUSTERS / 2) 418 #endif 419 #ifndef NMBUFS 420 #define NMBUFS (nmbclusters * 2 + maxfiles) 421 #endif 422 423 #define NMBCLUSTERS_MIN (NMBCLUSTERS / 2) 424 #define NMBJCLUSTERS_MIN (NMBJCLUSTERS / 2) 425 #define NMBUFS_MIN ((NMBCLUSTERS * 2 + maxfiles) / 2) 426 427 /* 428 * Perform sanity checks of tunables declared above. 429 */ 430 static void 431 tunable_mbinit(void *dummy) 432 { 433 /* 434 * This has to be done before VM init. 435 */ 436 nmbclusters = NMBCLUSTERS; 437 TUNABLE_INT_FETCH("kern.ipc.nmbclusters", &nmbclusters); 438 mjclph_cachefrac = MJCLPH_CACHEFRAC; 439 TUNABLE_INT_FETCH("kern.ipc.mjclph_cachefrac", &mjclph_cachefrac); 440 mjcl_cachefrac = MJCL_CACHEFRAC; 441 TUNABLE_INT_FETCH("kern.ipc.mjcl_cachefrac", &mjcl_cachefrac); 442 mclph_cachefrac = MCLPH_CACHEFRAC; 443 TUNABLE_INT_FETCH("kern.ipc.mclph_cachefrac", &mclph_cachefrac); 444 mcl_cachefrac = MCL_CACHEFRAC; 445 TUNABLE_INT_FETCH("kern.ipc.mcl_cachefrac", &mcl_cachefrac); 446 447 /* 448 * WARNING! each mcl cache feeds two mbuf caches, so the minimum 449 * cachefrac is 2. For safety, use 3. 450 */ 451 if (mjclph_cachefrac < 3) 452 mjclph_cachefrac = 3; 453 if (mjcl_cachefrac < 3) 454 mjcl_cachefrac = 3; 455 if (mclph_cachefrac < 3) 456 mclph_cachefrac = 3; 457 if (mcl_cachefrac < 3) 458 mcl_cachefrac = 3; 459 460 nmbjclusters = NMBJCLUSTERS; 461 TUNABLE_INT_FETCH("kern.ipc.nmbjclusters", &nmbjclusters); 462 463 nmbufs = NMBUFS; 464 TUNABLE_INT_FETCH("kern.ipc.nmbufs", &nmbufs); 465 466 /* Sanity checks */ 467 if (nmbufs < nmbclusters * 2) 468 nmbufs = nmbclusters * 2; 469 } 470 SYSINIT(tunable_mbinit, SI_BOOT1_TUNABLES, SI_ORDER_ANY, 471 tunable_mbinit, NULL); 472 473 static void 474 mbinclimit(int *limit, int inc, int minlim) 475 { 476 int new_limit; 477 478 lockmgr(&mbupdate_lk, LK_EXCLUSIVE); 479 480 new_limit = *limit + inc; 481 if (new_limit < minlim) 482 new_limit = minlim; 483 484 if (*limit != new_limit) { 485 *limit = new_limit; 486 mbupdatelimits(); 487 } 488 489 lockmgr(&mbupdate_lk, LK_RELEASE); 490 } 491 492 static int 493 mbsetlimit(int *limit, int new_limit, int minlim) 494 { 495 if (new_limit < minlim) 496 return EINVAL; 497 498 lockmgr(&mbupdate_lk, LK_EXCLUSIVE); 499 mbinclimit(limit, new_limit - *limit, minlim); 500 lockmgr(&mbupdate_lk, LK_RELEASE); 501 return 0; 502 } 503 504 static int 505 sysctl_mblimit(SYSCTL_HANDLER_ARGS, int *limit, int minlim) 506 { 507 int error, value; 508 509 value = *limit; 510 error = sysctl_handle_int(oidp, &value, 0, req); 511 if (error || req->newptr == NULL) 512 return error; 513 514 return mbsetlimit(limit, value, minlim); 515 } 516 517 /* 518 * Sysctl support to update nmbclusters, nmbjclusters, and nmbufs. 519 */ 520 static int 521 sysctl_nmbclusters(SYSCTL_HANDLER_ARGS) 522 { 523 return sysctl_mblimit(oidp, arg1, arg2, req, &nmbclusters, 524 NMBCLUSTERS_MIN); 525 } 526 527 static int 528 sysctl_nmbjclusters(SYSCTL_HANDLER_ARGS) 529 { 530 return sysctl_mblimit(oidp, arg1, arg2, req, &nmbjclusters, 531 NMBJCLUSTERS_MIN); 532 } 533 534 static int 535 sysctl_nmbufs(SYSCTL_HANDLER_ARGS) 536 { 537 return sysctl_mblimit(oidp, arg1, arg2, req, &nmbufs, NMBUFS_MIN); 538 } 539 540 void 541 mcl_inclimit(int inc) 542 { 543 mbinclimit(&nmbclusters, inc, NMBCLUSTERS_MIN); 544 } 545 546 void 547 mjcl_inclimit(int inc) 548 { 549 mbinclimit(&nmbjclusters, inc, NMBJCLUSTERS_MIN); 550 } 551 552 void 553 mb_inclimit(int inc) 554 { 555 mbinclimit(&nmbufs, inc, NMBUFS_MIN); 556 } 557 558 /* "number of clusters of pages" */ 559 #define NCL_INIT 1 560 561 #define NMB_INIT 16 562 563 /* 564 * The mbuf object cache only guarantees that m_next and m_nextpkt are 565 * NULL and that m_data points to the beginning of the data area. In 566 * particular, m_len and m_pkthdr.len are uninitialized. It is the 567 * responsibility of the caller to initialize those fields before use. 568 */ 569 static __inline boolean_t 570 mbuf_ctor(void *obj, void *private, int ocflags) 571 { 572 struct mbuf *m = obj; 573 574 m->m_next = NULL; 575 m->m_nextpkt = NULL; 576 m->m_data = m->m_dat; 577 m->m_flags = 0; 578 579 return (TRUE); 580 } 581 582 /* 583 * Initialize the mbuf and the packet header fields. 584 */ 585 static boolean_t 586 mbufphdr_ctor(void *obj, void *private, int ocflags) 587 { 588 struct mbuf *m = obj; 589 590 m->m_next = NULL; 591 m->m_nextpkt = NULL; 592 m->m_data = m->m_pktdat; 593 m->m_flags = M_PKTHDR | M_PHCACHE; 594 595 m->m_pkthdr.rcvif = NULL; /* eliminate XXX JH */ 596 SLIST_INIT(&m->m_pkthdr.tags); 597 m->m_pkthdr.csum_flags = 0; /* eliminate XXX JH */ 598 m->m_pkthdr.fw_flags = 0; /* eliminate XXX JH */ 599 600 return (TRUE); 601 } 602 603 /* 604 * A mbcluster object consists of 2K (MCLBYTES) cluster and a refcount. 605 */ 606 static boolean_t 607 mclmeta_ctor(void *obj, void *private, int ocflags) 608 { 609 struct mbcluster *cl = obj; 610 void *buf; 611 612 if (ocflags & M_NOWAIT) 613 buf = kmalloc(MCLBYTES, M_MBUFCL, M_NOWAIT | M_ZERO); 614 else 615 buf = kmalloc(MCLBYTES, M_MBUFCL, M_INTWAIT | M_ZERO); 616 if (buf == NULL) 617 return (FALSE); 618 cl->mcl_refs = 0; 619 cl->mcl_data = buf; 620 return (TRUE); 621 } 622 623 static boolean_t 624 mjclmeta_ctor(void *obj, void *private, int ocflags) 625 { 626 struct mbcluster *cl = obj; 627 void *buf; 628 629 if (ocflags & M_NOWAIT) 630 buf = kmalloc(MJUMPAGESIZE, M_MBUFCL, M_NOWAIT | M_ZERO); 631 else 632 buf = kmalloc(MJUMPAGESIZE, M_MBUFCL, M_INTWAIT | M_ZERO); 633 if (buf == NULL) 634 return (FALSE); 635 cl->mcl_refs = 0; 636 cl->mcl_data = buf; 637 return (TRUE); 638 } 639 640 static void 641 mclmeta_dtor(void *obj, void *private) 642 { 643 struct mbcluster *mcl = obj; 644 645 KKASSERT(mcl->mcl_refs == 0); 646 kfree(mcl->mcl_data, M_MBUFCL); 647 } 648 649 static void 650 linkjcluster(struct mbuf *m, struct mbcluster *cl, uint size) 651 { 652 /* 653 * Add the cluster to the mbuf. The caller will detect that the 654 * mbuf now has an attached cluster. 655 */ 656 m->m_ext.ext_arg = cl; 657 m->m_ext.ext_buf = cl->mcl_data; 658 m->m_ext.ext_ref = m_mclref; 659 if (size != MCLBYTES) 660 m->m_ext.ext_free = m_mjclfree; 661 else 662 m->m_ext.ext_free = m_mclfree; 663 m->m_ext.ext_size = size; 664 atomic_add_int(&cl->mcl_refs, 1); 665 666 m->m_data = m->m_ext.ext_buf; 667 m->m_flags |= M_EXT | M_EXT_CLUSTER; 668 } 669 670 static void 671 linkcluster(struct mbuf *m, struct mbcluster *cl) 672 { 673 linkjcluster(m, cl, MCLBYTES); 674 } 675 676 static boolean_t 677 mbufphdrcluster_ctor(void *obj, void *private, int ocflags) 678 { 679 struct mbuf *m = obj; 680 struct mbcluster *cl; 681 682 mbufphdr_ctor(obj, private, ocflags); 683 cl = objcache_get(mclmeta_cache, ocflags); 684 if (cl == NULL) { 685 ++mbstat[mycpu->gd_cpuid].m_drops; 686 return (FALSE); 687 } 688 m->m_flags |= M_CLCACHE; 689 linkcluster(m, cl); 690 return (TRUE); 691 } 692 693 static boolean_t 694 mbufphdrjcluster_ctor(void *obj, void *private, int ocflags) 695 { 696 struct mbuf *m = obj; 697 struct mbcluster *cl; 698 699 mbufphdr_ctor(obj, private, ocflags); 700 cl = objcache_get(mjclmeta_cache, ocflags); 701 if (cl == NULL) { 702 ++mbstat[mycpu->gd_cpuid].m_drops; 703 return (FALSE); 704 } 705 m->m_flags |= M_CLCACHE; 706 linkjcluster(m, cl, MJUMPAGESIZE); 707 return (TRUE); 708 } 709 710 static boolean_t 711 mbufcluster_ctor(void *obj, void *private, int ocflags) 712 { 713 struct mbuf *m = obj; 714 struct mbcluster *cl; 715 716 mbuf_ctor(obj, private, ocflags); 717 cl = objcache_get(mclmeta_cache, ocflags); 718 if (cl == NULL) { 719 ++mbstat[mycpu->gd_cpuid].m_drops; 720 return (FALSE); 721 } 722 m->m_flags |= M_CLCACHE; 723 linkcluster(m, cl); 724 return (TRUE); 725 } 726 727 static boolean_t 728 mbufjcluster_ctor(void *obj, void *private, int ocflags) 729 { 730 struct mbuf *m = obj; 731 struct mbcluster *cl; 732 733 mbuf_ctor(obj, private, ocflags); 734 cl = objcache_get(mjclmeta_cache, ocflags); 735 if (cl == NULL) { 736 ++mbstat[mycpu->gd_cpuid].m_drops; 737 return (FALSE); 738 } 739 m->m_flags |= M_CLCACHE; 740 linkjcluster(m, cl, MJUMPAGESIZE); 741 return (TRUE); 742 } 743 744 /* 745 * Used for both the cluster and cluster PHDR caches. 746 * 747 * The mbuf may have lost its cluster due to sharing, deal 748 * with the situation by checking M_EXT. 749 */ 750 static void 751 mbufcluster_dtor(void *obj, void *private) 752 { 753 struct mbuf *m = obj; 754 struct mbcluster *mcl; 755 756 if (m->m_flags & M_EXT) { 757 KKASSERT((m->m_flags & M_EXT_CLUSTER) != 0); 758 mcl = m->m_ext.ext_arg; 759 KKASSERT(mcl->mcl_refs == 1); 760 mcl->mcl_refs = 0; 761 if (m->m_flags & M_EXT && m->m_ext.ext_size != MCLBYTES) 762 objcache_put(mjclmeta_cache, mcl); 763 else 764 objcache_put(mclmeta_cache, mcl); 765 } 766 } 767 768 struct objcache_malloc_args mbuf_malloc_args = { MSIZE, M_MBUF }; 769 struct objcache_malloc_args mclmeta_malloc_args = 770 { sizeof(struct mbcluster), M_MCLMETA }; 771 772 /* ARGSUSED*/ 773 static void 774 mbinit(void *dummy) 775 { 776 int mb_limit, cl_limit, ncl_limit, jcl_limit; 777 int limit; 778 int i; 779 780 /* 781 * Initialize statistics 782 */ 783 for (i = 0; i < ncpus; i++) { 784 mbstat[i].m_msize = MSIZE; 785 mbstat[i].m_mclbytes = MCLBYTES; 786 mbstat[i].m_mjumpagesize = MJUMPAGESIZE; 787 mbstat[i].m_minclsize = MINCLSIZE; 788 mbstat[i].m_mlen = MLEN; 789 mbstat[i].m_mhlen = MHLEN; 790 } 791 792 /* 793 * Create object caches and save cluster limits, which will 794 * be used to adjust backing kmalloc pools' limit later. 795 */ 796 797 mb_limit = cl_limit = 0; 798 799 limit = nmbufs; 800 mbuf_cache = objcache_create("mbuf", 801 limit, nmbufs / 4, 802 mbuf_ctor, NULL, NULL, 803 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args); 804 mb_limit += limit; 805 806 limit = nmbufs; 807 mbufphdr_cache = objcache_create("mbuf pkt hdr", 808 limit, nmbufs / 4, 809 mbufphdr_ctor, NULL, NULL, 810 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args); 811 mb_limit += limit; 812 813 ncl_limit = nmbclusters; 814 mclmeta_cache = objcache_create("cluster mbuf", 815 ncl_limit, nmbclusters / 4, 816 mclmeta_ctor, mclmeta_dtor, NULL, 817 objcache_malloc_alloc, objcache_malloc_free, &mclmeta_malloc_args); 818 cl_limit += ncl_limit; 819 820 jcl_limit = nmbjclusters; 821 mjclmeta_cache = objcache_create("jcluster mbuf", 822 jcl_limit, nmbjclusters / 4, 823 mjclmeta_ctor, mclmeta_dtor, NULL, 824 objcache_malloc_alloc, objcache_malloc_free, &mclmeta_malloc_args); 825 cl_limit += jcl_limit; 826 827 limit = nmbclusters; 828 mbufcluster_cache = objcache_create("mbuf + cluster", 829 limit, nmbclusters / mcl_cachefrac, 830 mbufcluster_ctor, mbufcluster_dtor, NULL, 831 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args); 832 mb_limit += limit; 833 834 limit = nmbclusters; 835 mbufphdrcluster_cache = objcache_create("mbuf pkt hdr + cluster", 836 limit, nmbclusters / mclph_cachefrac, 837 mbufphdrcluster_ctor, mbufcluster_dtor, NULL, 838 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args); 839 mb_limit += limit; 840 841 limit = nmbjclusters; 842 mbufjcluster_cache = objcache_create("mbuf + jcluster", 843 limit, nmbjclusters / mjcl_cachefrac, 844 mbufjcluster_ctor, mbufcluster_dtor, NULL, 845 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args); 846 mb_limit += limit; 847 848 limit = nmbjclusters; 849 mbufphdrjcluster_cache = objcache_create("mbuf pkt hdr + jcluster", 850 limit, nmbjclusters / mjclph_cachefrac, 851 mbufphdrjcluster_ctor, mbufcluster_dtor, NULL, 852 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args); 853 mb_limit += limit; 854 855 /* 856 * Adjust backing kmalloc pools' limit 857 * 858 * NOTE: We raise the limit by another 1/8 to take the effect 859 * of loosememuse into account. 860 */ 861 cl_limit += cl_limit / 8; 862 kmalloc_raise_limit(mclmeta_malloc_args.mtype, 863 mclmeta_malloc_args.objsize * (size_t)cl_limit); 864 kmalloc_raise_limit(M_MBUFCL, 865 (MCLBYTES * (size_t)ncl_limit) + 866 (MJUMPAGESIZE * (size_t)jcl_limit)); 867 868 mb_limit += mb_limit / 8; 869 kmalloc_raise_limit(mbuf_malloc_args.mtype, 870 mbuf_malloc_args.objsize * (size_t)mb_limit); 871 } 872 873 /* 874 * Adjust mbuf limits after changes have been made 875 * 876 * Caller must hold mbupdate_lk 877 */ 878 static void 879 mbupdatelimits(void) 880 { 881 int mb_limit, cl_limit, ncl_limit, jcl_limit; 882 int limit; 883 884 KASSERT(lockstatus(&mbupdate_lk, curthread) != 0, 885 ("mbupdate_lk is not held")); 886 887 /* 888 * Figure out adjustments to object caches after nmbufs, nmbclusters, 889 * or nmbjclusters has been modified. 890 */ 891 mb_limit = cl_limit = 0; 892 893 limit = nmbufs; 894 objcache_set_cluster_limit(mbuf_cache, limit); 895 mb_limit += limit; 896 897 limit = nmbufs; 898 objcache_set_cluster_limit(mbufphdr_cache, limit); 899 mb_limit += limit; 900 901 ncl_limit = nmbclusters; 902 objcache_set_cluster_limit(mclmeta_cache, ncl_limit); 903 cl_limit += ncl_limit; 904 905 jcl_limit = nmbjclusters; 906 objcache_set_cluster_limit(mjclmeta_cache, jcl_limit); 907 cl_limit += jcl_limit; 908 909 limit = nmbclusters; 910 objcache_set_cluster_limit(mbufcluster_cache, limit); 911 mb_limit += limit; 912 913 limit = nmbclusters; 914 objcache_set_cluster_limit(mbufphdrcluster_cache, limit); 915 mb_limit += limit; 916 917 limit = nmbjclusters; 918 objcache_set_cluster_limit(mbufjcluster_cache, limit); 919 mb_limit += limit; 920 921 limit = nmbjclusters; 922 objcache_set_cluster_limit(mbufphdrjcluster_cache, limit); 923 mb_limit += limit; 924 925 /* 926 * Adjust backing kmalloc pools' limit 927 * 928 * NOTE: We raise the limit by another 1/8 to take the effect 929 * of loosememuse into account. 930 */ 931 cl_limit += cl_limit / 8; 932 kmalloc_raise_limit(mclmeta_malloc_args.mtype, 933 mclmeta_malloc_args.objsize * (size_t)cl_limit); 934 kmalloc_raise_limit(M_MBUFCL, 935 (MCLBYTES * (size_t)ncl_limit) + 936 (MJUMPAGESIZE * (size_t)jcl_limit)); 937 mb_limit += mb_limit / 8; 938 kmalloc_raise_limit(mbuf_malloc_args.mtype, 939 mbuf_malloc_args.objsize * (size_t)mb_limit); 940 } 941 942 /* 943 * Return the number of references to this mbuf's data. 0 is returned 944 * if the mbuf is not M_EXT, a reference count is returned if it is 945 * M_EXT | M_EXT_CLUSTER, and 99 is returned if it is a special M_EXT. 946 */ 947 int 948 m_sharecount(struct mbuf *m) 949 { 950 switch (m->m_flags & (M_EXT | M_EXT_CLUSTER)) { 951 case 0: 952 return (0); 953 case M_EXT: 954 return (99); 955 case M_EXT | M_EXT_CLUSTER: 956 return (((struct mbcluster *)m->m_ext.ext_arg)->mcl_refs); 957 } 958 /* NOTREACHED */ 959 return (0); /* to shut up compiler */ 960 } 961 962 /* 963 * change mbuf to new type 964 */ 965 void 966 m_chtype(struct mbuf *m, int type) 967 { 968 struct globaldata *gd = mycpu; 969 970 ++mbtypes[gd->gd_cpuid].stats[type]; 971 --mbtypes[gd->gd_cpuid].stats[m->m_type]; 972 m->m_type = type; 973 } 974 975 static void 976 m_reclaim(void) 977 { 978 struct domain *dp; 979 struct protosw *pr; 980 981 kprintf("Debug: m_reclaim() called\n"); 982 983 SLIST_FOREACH(dp, &domains, dom_next) { 984 for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) { 985 if (pr->pr_drain) 986 (*pr->pr_drain)(); 987 } 988 } 989 ++mbstat[mycpu->gd_cpuid].m_drain; 990 } 991 992 static __inline void 993 updatestats(struct mbuf *m, int type) 994 { 995 struct globaldata *gd = mycpu; 996 997 m->m_type = type; 998 mbuftrack(m); 999 #ifdef MBUF_DEBUG 1000 KASSERT(m->m_next == NULL, ("mbuf %p: bad m_next in get", m)); 1001 KASSERT(m->m_nextpkt == NULL, ("mbuf %p: bad m_nextpkt in get", m)); 1002 #endif 1003 1004 ++mbtypes[gd->gd_cpuid].stats[type]; 1005 ++mbstat[gd->gd_cpuid].m_mbufs; 1006 1007 } 1008 1009 /* 1010 * Allocate an mbuf. 1011 */ 1012 struct mbuf * 1013 m_get(int how, int type) 1014 { 1015 struct mbuf *m; 1016 int ntries = 0; 1017 int ocf = MB_OCFLAG(how); 1018 1019 retryonce: 1020 1021 m = objcache_get(mbuf_cache, ocf); 1022 1023 if (m == NULL) { 1024 if ((ocf & M_WAITOK) && ntries++ == 0) { 1025 struct objcache *reclaimlist[] = { 1026 mbufphdr_cache, 1027 mbufcluster_cache, 1028 mbufphdrcluster_cache, 1029 mbufjcluster_cache, 1030 mbufphdrjcluster_cache 1031 }; 1032 const int nreclaims = NELEM(reclaimlist); 1033 1034 if (!objcache_reclaimlist(reclaimlist, nreclaims, ocf)) 1035 m_reclaim(); 1036 goto retryonce; 1037 } 1038 ++mbstat[mycpu->gd_cpuid].m_drops; 1039 return (NULL); 1040 } 1041 #ifdef MBUF_DEBUG 1042 KASSERT(m->m_data == m->m_dat, ("mbuf %p: bad m_data in get", m)); 1043 #endif 1044 m->m_len = 0; 1045 1046 updatestats(m, type); 1047 return (m); 1048 } 1049 1050 struct mbuf * 1051 m_gethdr(int how, int type) 1052 { 1053 struct mbuf *m; 1054 int ocf = MB_OCFLAG(how); 1055 int ntries = 0; 1056 1057 retryonce: 1058 1059 m = objcache_get(mbufphdr_cache, ocf); 1060 1061 if (m == NULL) { 1062 if ((ocf & M_WAITOK) && ntries++ == 0) { 1063 struct objcache *reclaimlist[] = { 1064 mbuf_cache, 1065 mbufcluster_cache, mbufphdrcluster_cache, 1066 mbufjcluster_cache, mbufphdrjcluster_cache 1067 }; 1068 const int nreclaims = NELEM(reclaimlist); 1069 1070 if (!objcache_reclaimlist(reclaimlist, nreclaims, ocf)) 1071 m_reclaim(); 1072 goto retryonce; 1073 } 1074 ++mbstat[mycpu->gd_cpuid].m_drops; 1075 return (NULL); 1076 } 1077 #ifdef MBUF_DEBUG 1078 KASSERT(m->m_data == m->m_pktdat, ("mbuf %p: bad m_data in get", m)); 1079 #endif 1080 m->m_len = 0; 1081 m->m_pkthdr.len = 0; 1082 1083 updatestats(m, type); 1084 return (m); 1085 } 1086 1087 /* 1088 * Get a mbuf (not a mbuf cluster!) and zero it. 1089 * Deprecated. 1090 */ 1091 struct mbuf * 1092 m_getclr(int how, int type) 1093 { 1094 struct mbuf *m; 1095 1096 m = m_get(how, type); 1097 if (m != NULL) 1098 bzero(m->m_data, MLEN); 1099 return (m); 1100 } 1101 1102 static struct mbuf * 1103 m_getcl_cache(int how, short type, int flags, struct objcache *mbclc, 1104 struct objcache *mbphclc, u_long *cl_stats) 1105 { 1106 struct mbuf *m = NULL; 1107 int ocflags = MB_OCFLAG(how); 1108 int ntries = 0; 1109 1110 retryonce: 1111 1112 if (flags & M_PKTHDR) 1113 m = objcache_get(mbphclc, ocflags); 1114 else 1115 m = objcache_get(mbclc, ocflags); 1116 1117 if (m == NULL) { 1118 if ((ocflags & M_WAITOK) && ntries++ == 0) { 1119 struct objcache *reclaimlist[1]; 1120 1121 if (flags & M_PKTHDR) 1122 reclaimlist[0] = mbclc; 1123 else 1124 reclaimlist[0] = mbphclc; 1125 if (!objcache_reclaimlist(reclaimlist, 1, ocflags)) 1126 m_reclaim(); 1127 goto retryonce; 1128 } 1129 ++mbstat[mycpu->gd_cpuid].m_drops; 1130 return (NULL); 1131 } 1132 1133 #ifdef MBUF_DEBUG 1134 KASSERT(m->m_data == m->m_ext.ext_buf, 1135 ("mbuf %p: bad m_data in get", m)); 1136 #endif 1137 m->m_type = type; 1138 m->m_len = 0; 1139 m->m_pkthdr.len = 0; /* just do it unconditonally */ 1140 1141 mbuftrack(m); 1142 1143 ++mbtypes[mycpu->gd_cpuid].stats[type]; 1144 ++(*cl_stats); 1145 return (m); 1146 } 1147 1148 struct mbuf * 1149 m_getjcl(int how, short type, int flags, size_t size) 1150 { 1151 struct objcache *mbclc, *mbphclc; 1152 u_long *cl_stats; 1153 1154 switch (size) { 1155 case MCLBYTES: 1156 mbclc = mbufcluster_cache; 1157 mbphclc = mbufphdrcluster_cache; 1158 cl_stats = &mbstat[mycpu->gd_cpuid].m_clusters; 1159 break; 1160 1161 default: 1162 mbclc = mbufjcluster_cache; 1163 mbphclc = mbufphdrjcluster_cache; 1164 cl_stats = &mbstat[mycpu->gd_cpuid].m_jclusters; 1165 break; 1166 } 1167 return m_getcl_cache(how, type, flags, mbclc, mbphclc, cl_stats); 1168 } 1169 1170 /* 1171 * Returns an mbuf with an attached cluster. 1172 * Because many network drivers use this kind of buffers a lot, it is 1173 * convenient to keep a small pool of free buffers of this kind. 1174 * Even a small size such as 10 gives about 10% improvement in the 1175 * forwarding rate in a bridge or router. 1176 */ 1177 struct mbuf * 1178 m_getcl(int how, short type, int flags) 1179 { 1180 return m_getcl_cache(how, type, flags, 1181 mbufcluster_cache, mbufphdrcluster_cache, 1182 &mbstat[mycpu->gd_cpuid].m_clusters); 1183 } 1184 1185 /* 1186 * Allocate chain of requested length. 1187 */ 1188 struct mbuf * 1189 m_getc(int len, int how, int type) 1190 { 1191 struct mbuf *n, *nfirst = NULL, **ntail = &nfirst; 1192 int nsize; 1193 1194 while (len > 0) { 1195 n = m_getl(len, how, type, 0, &nsize); 1196 if (n == NULL) 1197 goto failed; 1198 n->m_len = 0; 1199 *ntail = n; 1200 ntail = &n->m_next; 1201 len -= nsize; 1202 } 1203 return (nfirst); 1204 1205 failed: 1206 m_freem(nfirst); 1207 return (NULL); 1208 } 1209 1210 /* 1211 * Allocate len-worth of mbufs and/or mbuf clusters (whatever fits best) 1212 * and return a pointer to the head of the allocated chain. If m0 is 1213 * non-null, then we assume that it is a single mbuf or an mbuf chain to 1214 * which we want len bytes worth of mbufs and/or clusters attached, and so 1215 * if we succeed in allocating it, we will just return a pointer to m0. 1216 * 1217 * If we happen to fail at any point during the allocation, we will free 1218 * up everything we have already allocated and return NULL. 1219 * 1220 * Deprecated. Use m_getc() and m_cat() instead. 1221 */ 1222 struct mbuf * 1223 m_getm(struct mbuf *m0, int len, int type, int how) 1224 { 1225 struct mbuf *nfirst; 1226 1227 nfirst = m_getc(len, how, type); 1228 1229 if (m0 != NULL) { 1230 m_last(m0)->m_next = nfirst; 1231 return (m0); 1232 } 1233 1234 return (nfirst); 1235 } 1236 1237 /* 1238 * Adds a cluster to a normal mbuf, M_EXT is set on success. 1239 * Deprecated. Use m_getcl() instead. 1240 */ 1241 void 1242 m_mclget(struct mbuf *m, int how) 1243 { 1244 struct mbcluster *mcl; 1245 1246 KKASSERT((m->m_flags & M_EXT) == 0); 1247 mcl = objcache_get(mclmeta_cache, MB_OCFLAG(how)); 1248 if (mcl != NULL) { 1249 linkcluster(m, mcl); 1250 ++mbstat[mycpu->gd_cpuid].m_clusters; 1251 } else { 1252 ++mbstat[mycpu->gd_cpuid].m_drops; 1253 } 1254 } 1255 1256 /* 1257 * Updates to mbcluster must be MPSAFE. Only an entity which already has 1258 * a reference to the cluster can ref it, so we are in no danger of 1259 * racing an add with a subtract. But the operation must still be atomic 1260 * since multiple entities may have a reference on the cluster. 1261 * 1262 * m_mclfree() is almost the same but it must contend with two entities 1263 * freeing the cluster at the same time. 1264 */ 1265 static void 1266 m_mclref(void *arg) 1267 { 1268 struct mbcluster *mcl = arg; 1269 1270 atomic_add_int(&mcl->mcl_refs, 1); 1271 } 1272 1273 /* 1274 * When dereferencing a cluster we have to deal with a N->0 race, where 1275 * N entities free their references simultaniously. To do this we use 1276 * atomic_fetchadd_int(). 1277 */ 1278 static void 1279 m_mclfree(void *arg) 1280 { 1281 struct mbcluster *mcl = arg; 1282 1283 if (atomic_fetchadd_int(&mcl->mcl_refs, -1) == 1) { 1284 --mbstat[mycpu->gd_cpuid].m_clusters; 1285 objcache_put(mclmeta_cache, mcl); 1286 } 1287 } 1288 1289 static void 1290 m_mjclfree(void *arg) 1291 { 1292 struct mbcluster *mcl = arg; 1293 1294 if (atomic_fetchadd_int(&mcl->mcl_refs, -1) == 1) { 1295 --mbstat[mycpu->gd_cpuid].m_jclusters; 1296 objcache_put(mjclmeta_cache, mcl); 1297 } 1298 } 1299 1300 /* 1301 * Free a single mbuf and any associated external storage. The successor, 1302 * if any, is returned. 1303 * 1304 * We do need to check non-first mbuf for m_aux, since some of existing 1305 * code does not call M_PREPEND properly. 1306 * (example: call to bpf_mtap from drivers) 1307 */ 1308 1309 #ifdef MBUF_DEBUG 1310 1311 struct mbuf * 1312 _m_free(struct mbuf *m, const char *func) 1313 1314 #else 1315 1316 struct mbuf * 1317 m_free(struct mbuf *m) 1318 1319 #endif 1320 { 1321 struct mbuf *n; 1322 struct globaldata *gd = mycpu; 1323 1324 KASSERT(m->m_type != MT_FREE, ("freeing free mbuf %p", m)); 1325 KASSERT(M_TRAILINGSPACE(m) >= 0, ("overflowed mbuf %p", m)); 1326 --mbtypes[gd->gd_cpuid].stats[m->m_type]; 1327 1328 n = m->m_next; 1329 1330 /* 1331 * Make sure the mbuf is in constructed state before returning it 1332 * to the objcache. 1333 */ 1334 m->m_next = NULL; 1335 mbufuntrack(m); 1336 #ifdef MBUF_DEBUG 1337 m->m_hdr.mh_lastfunc = func; 1338 #endif 1339 #ifdef notyet 1340 KKASSERT(m->m_nextpkt == NULL); 1341 #else 1342 if (m->m_nextpkt != NULL) { 1343 static int afewtimes = 10; 1344 1345 if (afewtimes-- > 0) { 1346 kprintf("mfree: m->m_nextpkt != NULL\n"); 1347 print_backtrace(-1); 1348 } 1349 m->m_nextpkt = NULL; 1350 } 1351 #endif 1352 if (m->m_flags & M_PKTHDR) { 1353 m_tag_delete_chain(m); /* eliminate XXX JH */ 1354 } 1355 1356 m->m_flags &= (M_EXT | M_EXT_CLUSTER | M_CLCACHE | M_PHCACHE); 1357 1358 /* 1359 * Clean the M_PKTHDR state so we can return the mbuf to its original 1360 * cache. This is based on the PHCACHE flag which tells us whether 1361 * the mbuf was originally allocated out of a packet-header cache 1362 * or a non-packet-header cache. 1363 */ 1364 if (m->m_flags & M_PHCACHE) { 1365 m->m_flags |= M_PKTHDR; 1366 m->m_pkthdr.rcvif = NULL; /* eliminate XXX JH */ 1367 m->m_pkthdr.csum_flags = 0; /* eliminate XXX JH */ 1368 m->m_pkthdr.fw_flags = 0; /* eliminate XXX JH */ 1369 SLIST_INIT(&m->m_pkthdr.tags); 1370 } 1371 1372 /* 1373 * Handle remaining flags combinations. M_CLCACHE tells us whether 1374 * the mbuf was originally allocated from a cluster cache or not, 1375 * and is totally separate from whether the mbuf is currently 1376 * associated with a cluster. 1377 */ 1378 switch(m->m_flags & (M_CLCACHE | M_EXT | M_EXT_CLUSTER)) { 1379 case M_CLCACHE | M_EXT | M_EXT_CLUSTER: 1380 /* 1381 * mbuf+cluster cache case. The mbuf was allocated from the 1382 * combined mbuf_cluster cache and can be returned to the 1383 * cache if the cluster hasn't been shared. 1384 */ 1385 if (m_sharecount(m) == 1) { 1386 /* 1387 * The cluster has not been shared, we can just 1388 * reset the data pointer and return the mbuf 1389 * to the cluster cache. Note that the reference 1390 * count is left intact (it is still associated with 1391 * an mbuf). 1392 */ 1393 m->m_data = m->m_ext.ext_buf; 1394 if (m->m_flags & M_EXT && m->m_ext.ext_size != MCLBYTES) { 1395 if (m->m_flags & M_PHCACHE) 1396 objcache_put(mbufphdrjcluster_cache, m); 1397 else 1398 objcache_put(mbufjcluster_cache, m); 1399 --mbstat[mycpu->gd_cpuid].m_jclusters; 1400 } else { 1401 if (m->m_flags & M_PHCACHE) 1402 objcache_put(mbufphdrcluster_cache, m); 1403 else 1404 objcache_put(mbufcluster_cache, m); 1405 --mbstat[mycpu->gd_cpuid].m_clusters; 1406 } 1407 } else { 1408 /* 1409 * Hell. Someone else has a ref on this cluster, 1410 * we have to disconnect it which means we can't 1411 * put it back into the mbufcluster_cache, we 1412 * have to destroy the mbuf. 1413 * 1414 * Other mbuf references to the cluster will typically 1415 * be M_EXT | M_EXT_CLUSTER but without M_CLCACHE. 1416 * 1417 * XXX we could try to connect another cluster to 1418 * it. 1419 */ 1420 m->m_ext.ext_free(m->m_ext.ext_arg); 1421 m->m_flags &= ~(M_EXT | M_EXT_CLUSTER); 1422 if (m->m_ext.ext_size == MCLBYTES) { 1423 if (m->m_flags & M_PHCACHE) 1424 objcache_dtor(mbufphdrcluster_cache, m); 1425 else 1426 objcache_dtor(mbufcluster_cache, m); 1427 } else { 1428 if (m->m_flags & M_PHCACHE) 1429 objcache_dtor(mbufphdrjcluster_cache, m); 1430 else 1431 objcache_dtor(mbufjcluster_cache, m); 1432 } 1433 } 1434 break; 1435 case M_EXT | M_EXT_CLUSTER: 1436 case M_EXT: 1437 /* 1438 * Normal cluster association case, disconnect the cluster from 1439 * the mbuf. The cluster may or may not be custom. 1440 */ 1441 m->m_ext.ext_free(m->m_ext.ext_arg); 1442 m->m_flags &= ~(M_EXT | M_EXT_CLUSTER); 1443 /* fall through */ 1444 case 0: 1445 /* 1446 * return the mbuf to the mbuf cache. 1447 */ 1448 if (m->m_flags & M_PHCACHE) { 1449 m->m_data = m->m_pktdat; 1450 objcache_put(mbufphdr_cache, m); 1451 } else { 1452 m->m_data = m->m_dat; 1453 objcache_put(mbuf_cache, m); 1454 } 1455 --mbstat[mycpu->gd_cpuid].m_mbufs; 1456 break; 1457 default: 1458 if (!panicstr) 1459 panic("bad mbuf flags %p %08x", m, m->m_flags); 1460 break; 1461 } 1462 return (n); 1463 } 1464 1465 #ifdef MBUF_DEBUG 1466 1467 void 1468 _m_freem(struct mbuf *m, const char *func) 1469 { 1470 while (m) 1471 m = _m_free(m, func); 1472 } 1473 1474 #else 1475 1476 void 1477 m_freem(struct mbuf *m) 1478 { 1479 while (m) 1480 m = m_free(m); 1481 } 1482 1483 #endif 1484 1485 void 1486 m_extadd(struct mbuf *m, caddr_t buf, u_int size, void (*reff)(void *), 1487 void (*freef)(void *), void *arg) 1488 { 1489 m->m_ext.ext_arg = arg; 1490 m->m_ext.ext_buf = buf; 1491 m->m_ext.ext_ref = reff; 1492 m->m_ext.ext_free = freef; 1493 m->m_ext.ext_size = size; 1494 reff(arg); 1495 m->m_data = buf; 1496 m->m_flags |= M_EXT; 1497 } 1498 1499 /* 1500 * mbuf utility routines 1501 */ 1502 1503 /* 1504 * Lesser-used path for M_PREPEND: allocate new mbuf to prepend to chain and 1505 * copy junk along. 1506 */ 1507 struct mbuf * 1508 m_prepend(struct mbuf *m, int len, int how) 1509 { 1510 struct mbuf *mn; 1511 1512 if (m->m_flags & M_PKTHDR) 1513 mn = m_gethdr(how, m->m_type); 1514 else 1515 mn = m_get(how, m->m_type); 1516 if (mn == NULL) { 1517 m_freem(m); 1518 return (NULL); 1519 } 1520 if (m->m_flags & M_PKTHDR) 1521 M_MOVE_PKTHDR(mn, m); 1522 mn->m_next = m; 1523 m = mn; 1524 if (len < MHLEN) 1525 MH_ALIGN(m, len); 1526 m->m_len = len; 1527 return (m); 1528 } 1529 1530 /* 1531 * Make a copy of an mbuf chain starting "off0" bytes from the beginning, 1532 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf. 1533 * The wait parameter is a choice of M_WAITOK/M_NOWAIT from caller. 1534 * Note that the copy is read-only, because clusters are not copied, 1535 * only their reference counts are incremented. 1536 */ 1537 struct mbuf * 1538 m_copym(const struct mbuf *m, int off0, int len, int wait) 1539 { 1540 struct mbuf *n, **np; 1541 int off = off0; 1542 struct mbuf *top; 1543 int copyhdr = 0; 1544 1545 KASSERT(off >= 0, ("m_copym, negative off %d", off)); 1546 KASSERT(len >= 0, ("m_copym, negative len %d", len)); 1547 if (off == 0 && (m->m_flags & M_PKTHDR)) 1548 copyhdr = 1; 1549 while (off > 0) { 1550 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain")); 1551 if (off < m->m_len) 1552 break; 1553 off -= m->m_len; 1554 m = m->m_next; 1555 } 1556 np = ⊤ 1557 top = NULL; 1558 while (len > 0) { 1559 if (m == NULL) { 1560 KASSERT(len == M_COPYALL, 1561 ("m_copym, length > size of mbuf chain")); 1562 break; 1563 } 1564 /* 1565 * Because we are sharing any cluster attachment below, 1566 * be sure to get an mbuf that does not have a cluster 1567 * associated with it. 1568 */ 1569 if (copyhdr) 1570 n = m_gethdr(wait, m->m_type); 1571 else 1572 n = m_get(wait, m->m_type); 1573 *np = n; 1574 if (n == NULL) 1575 goto nospace; 1576 if (copyhdr) { 1577 if (!m_dup_pkthdr(n, m, wait)) 1578 goto nospace; 1579 if (len == M_COPYALL) 1580 n->m_pkthdr.len -= off0; 1581 else 1582 n->m_pkthdr.len = len; 1583 copyhdr = 0; 1584 } 1585 n->m_len = min(len, m->m_len - off); 1586 if (m->m_flags & M_EXT) { 1587 KKASSERT((n->m_flags & M_EXT) == 0); 1588 n->m_data = m->m_data + off; 1589 m->m_ext.ext_ref(m->m_ext.ext_arg); 1590 n->m_ext = m->m_ext; 1591 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER); 1592 } else { 1593 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t), 1594 (unsigned)n->m_len); 1595 } 1596 if (len != M_COPYALL) 1597 len -= n->m_len; 1598 off = 0; 1599 m = m->m_next; 1600 np = &n->m_next; 1601 } 1602 if (top == NULL) 1603 ++mbstat[mycpu->gd_cpuid].m_mcfail; 1604 return (top); 1605 nospace: 1606 m_freem(top); 1607 ++mbstat[mycpu->gd_cpuid].m_mcfail; 1608 return (NULL); 1609 } 1610 1611 /* 1612 * Copy an entire packet, including header (which must be present). 1613 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'. 1614 * Note that the copy is read-only, because clusters are not copied, 1615 * only their reference counts are incremented. 1616 * Preserve alignment of the first mbuf so if the creator has left 1617 * some room at the beginning (e.g. for inserting protocol headers) 1618 * the copies also have the room available. 1619 */ 1620 struct mbuf * 1621 m_copypacket(struct mbuf *m, int how) 1622 { 1623 struct mbuf *top, *n, *o; 1624 1625 n = m_gethdr(how, m->m_type); 1626 top = n; 1627 if (!n) 1628 goto nospace; 1629 1630 if (!m_dup_pkthdr(n, m, how)) 1631 goto nospace; 1632 n->m_len = m->m_len; 1633 if (m->m_flags & M_EXT) { 1634 KKASSERT((n->m_flags & M_EXT) == 0); 1635 n->m_data = m->m_data; 1636 m->m_ext.ext_ref(m->m_ext.ext_arg); 1637 n->m_ext = m->m_ext; 1638 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER); 1639 } else { 1640 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat ); 1641 bcopy(mtod(m, char *), mtod(n, char *), n->m_len); 1642 } 1643 1644 m = m->m_next; 1645 while (m) { 1646 o = m_get(how, m->m_type); 1647 if (!o) 1648 goto nospace; 1649 1650 n->m_next = o; 1651 n = n->m_next; 1652 1653 n->m_len = m->m_len; 1654 if (m->m_flags & M_EXT) { 1655 KKASSERT((n->m_flags & M_EXT) == 0); 1656 n->m_data = m->m_data; 1657 m->m_ext.ext_ref(m->m_ext.ext_arg); 1658 n->m_ext = m->m_ext; 1659 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER); 1660 } else { 1661 bcopy(mtod(m, char *), mtod(n, char *), n->m_len); 1662 } 1663 1664 m = m->m_next; 1665 } 1666 return top; 1667 nospace: 1668 m_freem(top); 1669 ++mbstat[mycpu->gd_cpuid].m_mcfail; 1670 return (NULL); 1671 } 1672 1673 /* 1674 * Copy data from an mbuf chain starting "off" bytes from the beginning, 1675 * continuing for "len" bytes, into the indicated buffer. 1676 */ 1677 void 1678 m_copydata(const struct mbuf *m, int off, int len, caddr_t cp) 1679 { 1680 unsigned count; 1681 1682 KASSERT(off >= 0, ("m_copydata, negative off %d", off)); 1683 KASSERT(len >= 0, ("m_copydata, negative len %d", len)); 1684 while (off > 0) { 1685 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain")); 1686 if (off < m->m_len) 1687 break; 1688 off -= m->m_len; 1689 m = m->m_next; 1690 } 1691 while (len > 0) { 1692 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain")); 1693 count = min(m->m_len - off, len); 1694 bcopy(mtod(m, caddr_t) + off, cp, count); 1695 len -= count; 1696 cp += count; 1697 off = 0; 1698 m = m->m_next; 1699 } 1700 } 1701 1702 /* 1703 * Copy a packet header mbuf chain into a completely new chain, including 1704 * copying any mbuf clusters. Use this instead of m_copypacket() when 1705 * you need a writable copy of an mbuf chain. 1706 */ 1707 struct mbuf * 1708 m_dup(struct mbuf *m, int how) 1709 { 1710 struct mbuf **p, *top = NULL; 1711 int remain, moff, nsize; 1712 1713 /* Sanity check */ 1714 if (m == NULL) 1715 return (NULL); 1716 KASSERT((m->m_flags & M_PKTHDR) != 0, ("%s: !PKTHDR", __func__)); 1717 1718 /* While there's more data, get a new mbuf, tack it on, and fill it */ 1719 remain = m->m_pkthdr.len; 1720 moff = 0; 1721 p = ⊤ 1722 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */ 1723 struct mbuf *n; 1724 1725 /* Get the next new mbuf */ 1726 n = m_getl(remain, how, m->m_type, top == NULL ? M_PKTHDR : 0, 1727 &nsize); 1728 if (n == NULL) 1729 goto nospace; 1730 if (top == NULL) 1731 if (!m_dup_pkthdr(n, m, how)) 1732 goto nospace0; 1733 1734 /* Link it into the new chain */ 1735 *p = n; 1736 p = &n->m_next; 1737 1738 /* Copy data from original mbuf(s) into new mbuf */ 1739 n->m_len = 0; 1740 while (n->m_len < nsize && m != NULL) { 1741 int chunk = min(nsize - n->m_len, m->m_len - moff); 1742 1743 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk); 1744 moff += chunk; 1745 n->m_len += chunk; 1746 remain -= chunk; 1747 if (moff == m->m_len) { 1748 m = m->m_next; 1749 moff = 0; 1750 } 1751 } 1752 1753 /* Check correct total mbuf length */ 1754 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL), 1755 ("%s: bogus m_pkthdr.len", __func__)); 1756 } 1757 return (top); 1758 1759 nospace: 1760 m_freem(top); 1761 nospace0: 1762 ++mbstat[mycpu->gd_cpuid].m_mcfail; 1763 return (NULL); 1764 } 1765 1766 /* 1767 * Copy the non-packet mbuf data chain into a new set of mbufs, including 1768 * copying any mbuf clusters. This is typically used to realign a data 1769 * chain by nfs_realign(). 1770 * 1771 * The original chain is left intact. how should be M_WAITOK or M_NOWAIT 1772 * and NULL can be returned if M_NOWAIT is passed. 1773 * 1774 * Be careful to use cluster mbufs, a large mbuf chain converted to non 1775 * cluster mbufs can exhaust our supply of mbufs. 1776 */ 1777 struct mbuf * 1778 m_dup_data(struct mbuf *m, int how) 1779 { 1780 struct mbuf **p, *n, *top = NULL; 1781 int mlen, moff, chunk, gsize, nsize; 1782 1783 /* 1784 * Degenerate case 1785 */ 1786 if (m == NULL) 1787 return (NULL); 1788 1789 /* 1790 * Optimize the mbuf allocation but do not get too carried away. 1791 */ 1792 if (m->m_next || m->m_len > MLEN) 1793 if (m->m_flags & M_EXT && m->m_ext.ext_size == MCLBYTES) 1794 gsize = MCLBYTES; 1795 else 1796 gsize = MJUMPAGESIZE; 1797 else 1798 gsize = MLEN; 1799 1800 /* Chain control */ 1801 p = ⊤ 1802 n = NULL; 1803 nsize = 0; 1804 1805 /* 1806 * Scan the mbuf chain until nothing is left, the new mbuf chain 1807 * will be allocated on the fly as needed. 1808 */ 1809 while (m) { 1810 mlen = m->m_len; 1811 moff = 0; 1812 1813 while (mlen) { 1814 KKASSERT(m->m_type == MT_DATA); 1815 if (n == NULL) { 1816 n = m_getl(gsize, how, MT_DATA, 0, &nsize); 1817 n->m_len = 0; 1818 if (n == NULL) 1819 goto nospace; 1820 *p = n; 1821 p = &n->m_next; 1822 } 1823 chunk = imin(mlen, nsize); 1824 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk); 1825 mlen -= chunk; 1826 moff += chunk; 1827 n->m_len += chunk; 1828 nsize -= chunk; 1829 if (nsize == 0) 1830 n = NULL; 1831 } 1832 m = m->m_next; 1833 } 1834 *p = NULL; 1835 return(top); 1836 nospace: 1837 *p = NULL; 1838 m_freem(top); 1839 ++mbstat[mycpu->gd_cpuid].m_mcfail; 1840 return (NULL); 1841 } 1842 1843 /* 1844 * Concatenate mbuf chain n to m. 1845 * Both chains must be of the same type (e.g. MT_DATA). 1846 * Any m_pkthdr is not updated. 1847 */ 1848 void 1849 m_cat(struct mbuf *m, struct mbuf *n) 1850 { 1851 m = m_last(m); 1852 while (n) { 1853 if (m->m_flags & M_EXT || 1854 m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) { 1855 /* just join the two chains */ 1856 m->m_next = n; 1857 return; 1858 } 1859 /* splat the data from one into the other */ 1860 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 1861 (u_int)n->m_len); 1862 m->m_len += n->m_len; 1863 n = m_free(n); 1864 } 1865 } 1866 1867 void 1868 m_adj(struct mbuf *mp, int req_len) 1869 { 1870 int len = req_len; 1871 struct mbuf *m; 1872 int count; 1873 1874 if ((m = mp) == NULL) 1875 return; 1876 if (len >= 0) { 1877 /* 1878 * Trim from head. 1879 */ 1880 while (m != NULL && len > 0) { 1881 if (m->m_len <= len) { 1882 len -= m->m_len; 1883 m->m_len = 0; 1884 m = m->m_next; 1885 } else { 1886 m->m_len -= len; 1887 m->m_data += len; 1888 len = 0; 1889 } 1890 } 1891 m = mp; 1892 if (mp->m_flags & M_PKTHDR) 1893 m->m_pkthdr.len -= (req_len - len); 1894 } else { 1895 /* 1896 * Trim from tail. Scan the mbuf chain, 1897 * calculating its length and finding the last mbuf. 1898 * If the adjustment only affects this mbuf, then just 1899 * adjust and return. Otherwise, rescan and truncate 1900 * after the remaining size. 1901 */ 1902 len = -len; 1903 count = 0; 1904 for (;;) { 1905 count += m->m_len; 1906 if (m->m_next == NULL) 1907 break; 1908 m = m->m_next; 1909 } 1910 if (m->m_len >= len) { 1911 m->m_len -= len; 1912 if (mp->m_flags & M_PKTHDR) 1913 mp->m_pkthdr.len -= len; 1914 return; 1915 } 1916 count -= len; 1917 if (count < 0) 1918 count = 0; 1919 /* 1920 * Correct length for chain is "count". 1921 * Find the mbuf with last data, adjust its length, 1922 * and toss data from remaining mbufs on chain. 1923 */ 1924 m = mp; 1925 if (m->m_flags & M_PKTHDR) 1926 m->m_pkthdr.len = count; 1927 for (; m; m = m->m_next) { 1928 if (m->m_len >= count) { 1929 m->m_len = count; 1930 break; 1931 } 1932 count -= m->m_len; 1933 } 1934 while (m->m_next) 1935 (m = m->m_next) ->m_len = 0; 1936 } 1937 } 1938 1939 /* 1940 * Set the m_data pointer of a newly-allocated mbuf 1941 * to place an object of the specified size at the 1942 * end of the mbuf, longword aligned. 1943 */ 1944 void 1945 m_align(struct mbuf *m, int len) 1946 { 1947 int adjust; 1948 1949 if (m->m_flags & M_EXT) 1950 adjust = m->m_ext.ext_size - len; 1951 else if (m->m_flags & M_PKTHDR) 1952 adjust = MHLEN - len; 1953 else 1954 adjust = MLEN - len; 1955 m->m_data += adjust &~ (sizeof(long)-1); 1956 } 1957 1958 /* 1959 * Create a writable copy of the mbuf chain. While doing this 1960 * we compact the chain with a goal of producing a chain with 1961 * at most two mbufs. The second mbuf in this chain is likely 1962 * to be a cluster. The primary purpose of this work is to create 1963 * a writable packet for encryption, compression, etc. The 1964 * secondary goal is to linearize the data so the data can be 1965 * passed to crypto hardware in the most efficient manner possible. 1966 */ 1967 struct mbuf * 1968 m_unshare(struct mbuf *m0, int how) 1969 { 1970 struct mbuf *m, *mprev; 1971 struct mbuf *n, *mfirst, *mlast; 1972 int len, off; 1973 1974 mprev = NULL; 1975 for (m = m0; m != NULL; m = mprev->m_next) { 1976 /* 1977 * Regular mbufs are ignored unless there's a cluster 1978 * in front of it that we can use to coalesce. We do 1979 * the latter mainly so later clusters can be coalesced 1980 * also w/o having to handle them specially (i.e. convert 1981 * mbuf+cluster -> cluster). This optimization is heavily 1982 * influenced by the assumption that we're running over 1983 * Ethernet where MCLBYTES is large enough that the max 1984 * packet size will permit lots of coalescing into a 1985 * single cluster. This in turn permits efficient 1986 * crypto operations, especially when using hardware. 1987 */ 1988 if ((m->m_flags & M_EXT) == 0) { 1989 if (mprev && (mprev->m_flags & M_EXT) && 1990 m->m_len <= M_TRAILINGSPACE(mprev)) { 1991 /* XXX: this ignores mbuf types */ 1992 memcpy(mtod(mprev, caddr_t) + mprev->m_len, 1993 mtod(m, caddr_t), m->m_len); 1994 mprev->m_len += m->m_len; 1995 mprev->m_next = m->m_next; /* unlink from chain */ 1996 m_free(m); /* reclaim mbuf */ 1997 } else { 1998 mprev = m; 1999 } 2000 continue; 2001 } 2002 /* 2003 * Writable mbufs are left alone (for now). 2004 */ 2005 if (M_WRITABLE(m)) { 2006 mprev = m; 2007 continue; 2008 } 2009 2010 /* 2011 * Not writable, replace with a copy or coalesce with 2012 * the previous mbuf if possible (since we have to copy 2013 * it anyway, we try to reduce the number of mbufs and 2014 * clusters so that future work is easier). 2015 */ 2016 KASSERT(m->m_flags & M_EXT, ("m_flags 0x%x", m->m_flags)); 2017 /* NB: we only coalesce into a cluster or larger */ 2018 if (mprev != NULL && (mprev->m_flags & M_EXT) && 2019 m->m_len <= M_TRAILINGSPACE(mprev)) { 2020 /* XXX: this ignores mbuf types */ 2021 memcpy(mtod(mprev, caddr_t) + mprev->m_len, 2022 mtod(m, caddr_t), m->m_len); 2023 mprev->m_len += m->m_len; 2024 mprev->m_next = m->m_next; /* unlink from chain */ 2025 m_free(m); /* reclaim mbuf */ 2026 continue; 2027 } 2028 2029 /* 2030 * Allocate new space to hold the copy... 2031 */ 2032 /* XXX why can M_PKTHDR be set past the first mbuf? */ 2033 if (mprev == NULL && (m->m_flags & M_PKTHDR)) { 2034 /* 2035 * NB: if a packet header is present we must 2036 * allocate the mbuf separately from any cluster 2037 * because M_MOVE_PKTHDR will smash the data 2038 * pointer and drop the M_EXT marker. 2039 */ 2040 MGETHDR(n, how, m->m_type); 2041 if (n == NULL) { 2042 m_freem(m0); 2043 return (NULL); 2044 } 2045 M_MOVE_PKTHDR(n, m); 2046 MCLGET(n, how); 2047 if ((n->m_flags & M_EXT) == 0) { 2048 m_free(n); 2049 m_freem(m0); 2050 return (NULL); 2051 } 2052 } else { 2053 n = m_getcl(how, m->m_type, m->m_flags); 2054 if (n == NULL) { 2055 m_freem(m0); 2056 return (NULL); 2057 } 2058 } 2059 /* 2060 * ... and copy the data. We deal with jumbo mbufs 2061 * (i.e. m_len > MCLBYTES) by splitting them into 2062 * clusters. We could just malloc a buffer and make 2063 * it external but too many device drivers don't know 2064 * how to break up the non-contiguous memory when 2065 * doing DMA. 2066 */ 2067 len = m->m_len; 2068 off = 0; 2069 mfirst = n; 2070 mlast = NULL; 2071 for (;;) { 2072 int cc = min(len, MCLBYTES); 2073 memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + off, cc); 2074 n->m_len = cc; 2075 if (mlast != NULL) 2076 mlast->m_next = n; 2077 mlast = n; 2078 2079 len -= cc; 2080 if (len <= 0) 2081 break; 2082 off += cc; 2083 2084 n = m_getcl(how, m->m_type, m->m_flags); 2085 if (n == NULL) { 2086 m_freem(mfirst); 2087 m_freem(m0); 2088 return (NULL); 2089 } 2090 } 2091 n->m_next = m->m_next; 2092 if (mprev == NULL) 2093 m0 = mfirst; /* new head of chain */ 2094 else 2095 mprev->m_next = mfirst; /* replace old mbuf */ 2096 m_free(m); /* release old mbuf */ 2097 mprev = mfirst; 2098 } 2099 return (m0); 2100 } 2101 2102 /* 2103 * Rearrange an mbuf chain so that len bytes are contiguous 2104 * and in the data area of an mbuf (so that mtod will work for a structure 2105 * of size len). Returns the resulting mbuf chain on success, frees it and 2106 * returns null on failure. If there is room, it will add up to 2107 * max_protohdr-len extra bytes to the contiguous region in an attempt to 2108 * avoid being called next time. 2109 */ 2110 struct mbuf * 2111 m_pullup(struct mbuf *n, int len) 2112 { 2113 struct mbuf *m; 2114 int count; 2115 int space; 2116 2117 /* 2118 * If first mbuf has no cluster, and has room for len bytes 2119 * without shifting current data, pullup into it, 2120 * otherwise allocate a new mbuf to prepend to the chain. 2121 */ 2122 if (!(n->m_flags & M_EXT) && 2123 n->m_data + len < &n->m_dat[MLEN] && 2124 n->m_next) { 2125 if (n->m_len >= len) 2126 return (n); 2127 m = n; 2128 n = n->m_next; 2129 len -= m->m_len; 2130 } else { 2131 if (len > MHLEN) 2132 goto bad; 2133 if (n->m_flags & M_PKTHDR) 2134 m = m_gethdr(M_NOWAIT, n->m_type); 2135 else 2136 m = m_get(M_NOWAIT, n->m_type); 2137 if (m == NULL) 2138 goto bad; 2139 m->m_len = 0; 2140 if (n->m_flags & M_PKTHDR) 2141 M_MOVE_PKTHDR(m, n); 2142 } 2143 space = &m->m_dat[MLEN] - (m->m_data + m->m_len); 2144 do { 2145 count = min(min(max(len, max_protohdr), space), n->m_len); 2146 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 2147 (unsigned)count); 2148 len -= count; 2149 m->m_len += count; 2150 n->m_len -= count; 2151 space -= count; 2152 if (n->m_len) 2153 n->m_data += count; 2154 else 2155 n = m_free(n); 2156 } while (len > 0 && n); 2157 if (len > 0) { 2158 m_free(m); 2159 goto bad; 2160 } 2161 m->m_next = n; 2162 return (m); 2163 bad: 2164 m_freem(n); 2165 ++mbstat[mycpu->gd_cpuid].m_mcfail; 2166 return (NULL); 2167 } 2168 2169 /* 2170 * Partition an mbuf chain in two pieces, returning the tail -- 2171 * all but the first len0 bytes. In case of failure, it returns NULL and 2172 * attempts to restore the chain to its original state. 2173 * 2174 * Note that the resulting mbufs might be read-only, because the new 2175 * mbuf can end up sharing an mbuf cluster with the original mbuf if 2176 * the "breaking point" happens to lie within a cluster mbuf. Use the 2177 * M_WRITABLE() macro to check for this case. 2178 */ 2179 struct mbuf * 2180 m_split(struct mbuf *m0, int len0, int wait) 2181 { 2182 struct mbuf *m, *n; 2183 unsigned len = len0, remain; 2184 2185 for (m = m0; m && len > m->m_len; m = m->m_next) 2186 len -= m->m_len; 2187 if (m == NULL) 2188 return (NULL); 2189 remain = m->m_len - len; 2190 if (m0->m_flags & M_PKTHDR) { 2191 n = m_gethdr(wait, m0->m_type); 2192 if (n == NULL) 2193 return (NULL); 2194 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif; 2195 n->m_pkthdr.len = m0->m_pkthdr.len - len0; 2196 m0->m_pkthdr.len = len0; 2197 if (m->m_flags & M_EXT) 2198 goto extpacket; 2199 if (remain > MHLEN) { 2200 /* m can't be the lead packet */ 2201 MH_ALIGN(n, 0); 2202 n->m_next = m_split(m, len, wait); 2203 if (n->m_next == NULL) { 2204 m_free(n); 2205 return (NULL); 2206 } else { 2207 n->m_len = 0; 2208 return (n); 2209 } 2210 } else 2211 MH_ALIGN(n, remain); 2212 } else if (remain == 0) { 2213 n = m->m_next; 2214 m->m_next = NULL; 2215 return (n); 2216 } else { 2217 n = m_get(wait, m->m_type); 2218 if (n == NULL) 2219 return (NULL); 2220 M_ALIGN(n, remain); 2221 } 2222 extpacket: 2223 if (m->m_flags & M_EXT) { 2224 KKASSERT((n->m_flags & M_EXT) == 0); 2225 n->m_data = m->m_data + len; 2226 m->m_ext.ext_ref(m->m_ext.ext_arg); 2227 n->m_ext = m->m_ext; 2228 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER); 2229 } else { 2230 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain); 2231 } 2232 n->m_len = remain; 2233 m->m_len = len; 2234 n->m_next = m->m_next; 2235 m->m_next = NULL; 2236 return (n); 2237 } 2238 2239 /* 2240 * Routine to copy from device local memory into mbufs. 2241 * Note: "offset" is ill-defined and always called as 0, so ignore it. 2242 */ 2243 struct mbuf * 2244 m_devget(char *buf, int len, int offset, struct ifnet *ifp) 2245 { 2246 struct mbuf *m, *mfirst = NULL, **mtail; 2247 int nsize, flags; 2248 2249 mtail = &mfirst; 2250 flags = M_PKTHDR; 2251 2252 while (len > 0) { 2253 m = m_getl(len, M_NOWAIT, MT_DATA, flags, &nsize); 2254 if (m == NULL) { 2255 m_freem(mfirst); 2256 return (NULL); 2257 } 2258 m->m_len = min(len, nsize); 2259 2260 if (flags & M_PKTHDR) { 2261 if (len + max_linkhdr <= nsize) 2262 m->m_data += max_linkhdr; 2263 m->m_pkthdr.rcvif = ifp; 2264 m->m_pkthdr.len = len; 2265 flags = 0; 2266 } 2267 2268 bcopy(buf, m->m_data, (unsigned)m->m_len); 2269 buf += m->m_len; 2270 len -= m->m_len; 2271 *mtail = m; 2272 mtail = &m->m_next; 2273 } 2274 2275 return (mfirst); 2276 } 2277 2278 /* 2279 * Routine to pad mbuf to the specified length 'padto'. 2280 */ 2281 int 2282 m_devpad(struct mbuf *m, int padto) 2283 { 2284 struct mbuf *last = NULL; 2285 int padlen; 2286 2287 if (padto <= m->m_pkthdr.len) 2288 return 0; 2289 2290 padlen = padto - m->m_pkthdr.len; 2291 2292 /* if there's only the packet-header and we can pad there, use it. */ 2293 if (m->m_pkthdr.len == m->m_len && M_TRAILINGSPACE(m) >= padlen) { 2294 last = m; 2295 } else { 2296 /* 2297 * Walk packet chain to find last mbuf. We will either 2298 * pad there, or append a new mbuf and pad it 2299 */ 2300 for (last = m; last->m_next != NULL; last = last->m_next) 2301 ; /* EMPTY */ 2302 2303 /* `last' now points to last in chain. */ 2304 if (M_TRAILINGSPACE(last) < padlen) { 2305 struct mbuf *n; 2306 2307 /* Allocate new empty mbuf, pad it. Compact later. */ 2308 MGET(n, M_NOWAIT, MT_DATA); 2309 if (n == NULL) 2310 return ENOBUFS; 2311 n->m_len = 0; 2312 last->m_next = n; 2313 last = n; 2314 } 2315 } 2316 KKASSERT(M_TRAILINGSPACE(last) >= padlen); 2317 KKASSERT(M_WRITABLE(last)); 2318 2319 /* Now zero the pad area */ 2320 bzero(mtod(last, char *) + last->m_len, padlen); 2321 last->m_len += padlen; 2322 m->m_pkthdr.len += padlen; 2323 return 0; 2324 } 2325 2326 /* 2327 * Copy data from a buffer back into the indicated mbuf chain, 2328 * starting "off" bytes from the beginning, extending the mbuf 2329 * chain if necessary. 2330 */ 2331 void 2332 m_copyback(struct mbuf *m0, int off, int len, caddr_t cp) 2333 { 2334 int mlen; 2335 struct mbuf *m = m0, *n; 2336 int totlen = 0; 2337 2338 if (m0 == NULL) 2339 return; 2340 while (off > (mlen = m->m_len)) { 2341 off -= mlen; 2342 totlen += mlen; 2343 if (m->m_next == NULL) { 2344 n = m_getclr(M_NOWAIT, m->m_type); 2345 if (n == NULL) 2346 goto out; 2347 n->m_len = min(MLEN, len + off); 2348 m->m_next = n; 2349 } 2350 m = m->m_next; 2351 } 2352 while (len > 0) { 2353 mlen = min (m->m_len - off, len); 2354 bcopy(cp, off + mtod(m, caddr_t), (unsigned)mlen); 2355 cp += mlen; 2356 len -= mlen; 2357 mlen += off; 2358 off = 0; 2359 totlen += mlen; 2360 if (len == 0) 2361 break; 2362 if (m->m_next == NULL) { 2363 n = m_get(M_NOWAIT, m->m_type); 2364 if (n == NULL) 2365 break; 2366 n->m_len = min(MLEN, len); 2367 m->m_next = n; 2368 } 2369 m = m->m_next; 2370 } 2371 out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen)) 2372 m->m_pkthdr.len = totlen; 2373 } 2374 2375 /* 2376 * Append the specified data to the indicated mbuf chain, 2377 * Extend the mbuf chain if the new data does not fit in 2378 * existing space. 2379 * 2380 * Return 1 if able to complete the job; otherwise 0. 2381 */ 2382 int 2383 m_append(struct mbuf *m0, int len, c_caddr_t cp) 2384 { 2385 struct mbuf *m, *n; 2386 int remainder, space; 2387 2388 for (m = m0; m->m_next != NULL; m = m->m_next) 2389 ; 2390 remainder = len; 2391 space = M_TRAILINGSPACE(m); 2392 if (space > 0) { 2393 /* 2394 * Copy into available space. 2395 */ 2396 if (space > remainder) 2397 space = remainder; 2398 bcopy(cp, mtod(m, caddr_t) + m->m_len, space); 2399 m->m_len += space; 2400 cp += space, remainder -= space; 2401 } 2402 while (remainder > 0) { 2403 /* 2404 * Allocate a new mbuf; could check space 2405 * and allocate a cluster instead. 2406 */ 2407 n = m_get(M_NOWAIT, m->m_type); 2408 if (n == NULL) 2409 break; 2410 n->m_len = min(MLEN, remainder); 2411 bcopy(cp, mtod(n, caddr_t), n->m_len); 2412 cp += n->m_len, remainder -= n->m_len; 2413 m->m_next = n; 2414 m = n; 2415 } 2416 if (m0->m_flags & M_PKTHDR) 2417 m0->m_pkthdr.len += len - remainder; 2418 return (remainder == 0); 2419 } 2420 2421 /* 2422 * Apply function f to the data in an mbuf chain starting "off" bytes from 2423 * the beginning, continuing for "len" bytes. 2424 */ 2425 int 2426 m_apply(struct mbuf *m, int off, int len, 2427 int (*f)(void *, void *, u_int), void *arg) 2428 { 2429 u_int count; 2430 int rval; 2431 2432 KASSERT(off >= 0, ("m_apply, negative off %d", off)); 2433 KASSERT(len >= 0, ("m_apply, negative len %d", len)); 2434 while (off > 0) { 2435 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain")); 2436 if (off < m->m_len) 2437 break; 2438 off -= m->m_len; 2439 m = m->m_next; 2440 } 2441 while (len > 0) { 2442 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain")); 2443 count = min(m->m_len - off, len); 2444 rval = (*f)(arg, mtod(m, caddr_t) + off, count); 2445 if (rval) 2446 return (rval); 2447 len -= count; 2448 off = 0; 2449 m = m->m_next; 2450 } 2451 return (0); 2452 } 2453 2454 /* 2455 * Return a pointer to mbuf/offset of location in mbuf chain. 2456 */ 2457 struct mbuf * 2458 m_getptr(struct mbuf *m, int loc, int *off) 2459 { 2460 2461 while (loc >= 0) { 2462 /* Normal end of search. */ 2463 if (m->m_len > loc) { 2464 *off = loc; 2465 return (m); 2466 } else { 2467 loc -= m->m_len; 2468 if (m->m_next == NULL) { 2469 if (loc == 0) { 2470 /* Point at the end of valid data. */ 2471 *off = m->m_len; 2472 return (m); 2473 } 2474 return (NULL); 2475 } 2476 m = m->m_next; 2477 } 2478 } 2479 return (NULL); 2480 } 2481 2482 void 2483 m_print(const struct mbuf *m) 2484 { 2485 int len; 2486 const struct mbuf *m2; 2487 char *hexstr; 2488 2489 len = m->m_pkthdr.len; 2490 m2 = m; 2491 hexstr = kmalloc(HEX_NCPYLEN(len), M_TEMP, M_ZERO | M_WAITOK); 2492 while (len) { 2493 kprintf("%p %s\n", m2, hexncpy(m2->m_data, m2->m_len, hexstr, 2494 HEX_NCPYLEN(m2->m_len), "-")); 2495 len -= m2->m_len; 2496 m2 = m2->m_next; 2497 } 2498 kfree(hexstr, M_TEMP); 2499 return; 2500 } 2501 2502 /* 2503 * "Move" mbuf pkthdr from "from" to "to". 2504 * "from" must have M_PKTHDR set, and "to" must be empty. 2505 */ 2506 void 2507 m_move_pkthdr(struct mbuf *to, struct mbuf *from) 2508 { 2509 KASSERT((to->m_flags & M_PKTHDR), ("m_move_pkthdr: not packet header")); 2510 2511 to->m_flags |= from->m_flags & M_COPYFLAGS; 2512 to->m_pkthdr = from->m_pkthdr; /* especially tags */ 2513 SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */ 2514 } 2515 2516 /* 2517 * Duplicate "from"'s mbuf pkthdr in "to". 2518 * "from" must have M_PKTHDR set, and "to" must be empty. 2519 * In particular, this does a deep copy of the packet tags. 2520 */ 2521 int 2522 m_dup_pkthdr(struct mbuf *to, const struct mbuf *from, int how) 2523 { 2524 KASSERT((to->m_flags & M_PKTHDR), ("m_dup_pkthdr: not packet header")); 2525 2526 to->m_flags = (from->m_flags & M_COPYFLAGS) | 2527 (to->m_flags & ~M_COPYFLAGS); 2528 to->m_pkthdr = from->m_pkthdr; 2529 SLIST_INIT(&to->m_pkthdr.tags); 2530 return (m_tag_copy_chain(to, from, how)); 2531 } 2532 2533 /* 2534 * Defragment a mbuf chain, returning the shortest possible 2535 * chain of mbufs and clusters. If allocation fails and 2536 * this cannot be completed, NULL will be returned, but 2537 * the passed in chain will be unchanged. Upon success, 2538 * the original chain will be freed, and the new chain 2539 * will be returned. 2540 * 2541 * If a non-packet header is passed in, the original 2542 * mbuf (chain?) will be returned unharmed. 2543 * 2544 * m_defrag_nofree doesn't free the passed in mbuf. 2545 */ 2546 struct mbuf * 2547 m_defrag(struct mbuf *m0, int how) 2548 { 2549 struct mbuf *m_new; 2550 2551 if ((m_new = m_defrag_nofree(m0, how)) == NULL) 2552 return (NULL); 2553 if (m_new != m0) 2554 m_freem(m0); 2555 return (m_new); 2556 } 2557 2558 struct mbuf * 2559 m_defrag_nofree(struct mbuf *m0, int how) 2560 { 2561 struct mbuf *m_new = NULL, *m_final = NULL; 2562 int progress = 0, length, nsize; 2563 2564 if (!(m0->m_flags & M_PKTHDR)) 2565 return (m0); 2566 2567 #ifdef MBUF_STRESS_TEST 2568 if (m_defragrandomfailures) { 2569 int temp = karc4random() & 0xff; 2570 if (temp == 0xba) 2571 goto nospace; 2572 } 2573 #endif 2574 2575 m_final = m_getl(m0->m_pkthdr.len, how, MT_DATA, M_PKTHDR, &nsize); 2576 if (m_final == NULL) 2577 goto nospace; 2578 m_final->m_len = 0; /* in case m0->m_pkthdr.len is zero */ 2579 2580 if (m_dup_pkthdr(m_final, m0, how) == 0) 2581 goto nospace; 2582 2583 m_new = m_final; 2584 2585 while (progress < m0->m_pkthdr.len) { 2586 length = m0->m_pkthdr.len - progress; 2587 if (length > MCLBYTES) 2588 length = MCLBYTES; 2589 2590 if (m_new == NULL) { 2591 m_new = m_getl(length, how, MT_DATA, 0, &nsize); 2592 if (m_new == NULL) 2593 goto nospace; 2594 } 2595 2596 m_copydata(m0, progress, length, mtod(m_new, caddr_t)); 2597 progress += length; 2598 m_new->m_len = length; 2599 if (m_new != m_final) 2600 m_cat(m_final, m_new); 2601 m_new = NULL; 2602 } 2603 if (m0->m_next == NULL) 2604 m_defraguseless++; 2605 m_defragpackets++; 2606 m_defragbytes += m_final->m_pkthdr.len; 2607 return (m_final); 2608 nospace: 2609 m_defragfailure++; 2610 if (m_new) 2611 m_free(m_new); 2612 m_freem(m_final); 2613 return (NULL); 2614 } 2615 2616 /* 2617 * Move data from uio into mbufs. 2618 */ 2619 struct mbuf * 2620 m_uiomove(struct uio *uio) 2621 { 2622 struct mbuf *m; /* current working mbuf */ 2623 struct mbuf *head = NULL; /* result mbuf chain */ 2624 struct mbuf **mp = &head; 2625 int flags = M_PKTHDR; 2626 int nsize; 2627 int error; 2628 int resid; 2629 2630 do { 2631 if (uio->uio_resid > INT_MAX) 2632 resid = INT_MAX; 2633 else 2634 resid = (int)uio->uio_resid; 2635 m = m_getl(resid, M_WAITOK, MT_DATA, flags, &nsize); 2636 if (flags) { 2637 m->m_pkthdr.len = 0; 2638 /* Leave room for protocol headers. */ 2639 if (resid < MHLEN) 2640 MH_ALIGN(m, resid); 2641 flags = 0; 2642 } 2643 m->m_len = imin(nsize, resid); 2644 error = uiomove(mtod(m, caddr_t), m->m_len, uio); 2645 if (error) { 2646 m_free(m); 2647 goto failed; 2648 } 2649 *mp = m; 2650 mp = &m->m_next; 2651 head->m_pkthdr.len += m->m_len; 2652 } while (uio->uio_resid > 0); 2653 2654 return (head); 2655 2656 failed: 2657 m_freem(head); 2658 return (NULL); 2659 } 2660 2661 struct mbuf * 2662 m_last(struct mbuf *m) 2663 { 2664 while (m->m_next) 2665 m = m->m_next; 2666 return (m); 2667 } 2668 2669 /* 2670 * Return the number of bytes in an mbuf chain. 2671 * If lastm is not NULL, also return the last mbuf. 2672 */ 2673 u_int 2674 m_lengthm(struct mbuf *m, struct mbuf **lastm) 2675 { 2676 u_int len = 0; 2677 struct mbuf *prev = m; 2678 2679 while (m) { 2680 len += m->m_len; 2681 prev = m; 2682 m = m->m_next; 2683 } 2684 if (lastm != NULL) 2685 *lastm = prev; 2686 return (len); 2687 } 2688 2689 /* 2690 * Like m_lengthm(), except also keep track of mbuf usage. 2691 */ 2692 u_int 2693 m_countm(struct mbuf *m, struct mbuf **lastm, u_int *pmbcnt) 2694 { 2695 u_int len = 0, mbcnt = 0; 2696 struct mbuf *prev = m; 2697 2698 while (m) { 2699 len += m->m_len; 2700 mbcnt += MSIZE; 2701 if (m->m_flags & M_EXT) 2702 mbcnt += m->m_ext.ext_size; 2703 prev = m; 2704 m = m->m_next; 2705 } 2706 if (lastm != NULL) 2707 *lastm = prev; 2708 *pmbcnt = mbcnt; 2709 return (len); 2710 } 2711