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