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/proc.h> 84 #include <sys/globaldata.h> 85 86 #include <sys/thread2.h> 87 #include <sys/spinlock2.h> 88 89 #include <machine/atomic.h> 90 #include <machine/limits.h> 91 92 #include <vm/vm.h> 93 #include <vm/vm_kern.h> 94 #include <vm/vm_extern.h> 95 96 #ifdef INVARIANTS 97 #include <machine/cpu.h> 98 #endif 99 100 /* 101 * mbuf cluster meta-data 102 */ 103 struct mbcluster { 104 int32_t mcl_refs; 105 void *mcl_data; 106 }; 107 108 /* 109 * mbuf tracking for debugging purposes 110 */ 111 #ifdef MBUF_DEBUG 112 113 static MALLOC_DEFINE(M_MTRACK, "mtrack", "mtrack"); 114 115 struct mbctrack; 116 RB_HEAD(mbuf_rb_tree, mbtrack); 117 RB_PROTOTYPE2(mbuf_rb_tree, mbtrack, rb_node, mbtrack_cmp, struct mbuf *); 118 119 struct mbtrack { 120 RB_ENTRY(mbtrack) rb_node; 121 int trackid; 122 struct mbuf *m; 123 }; 124 125 static int 126 mbtrack_cmp(struct mbtrack *mb1, struct mbtrack *mb2) 127 { 128 if (mb1->m < mb2->m) 129 return(-1); 130 if (mb1->m > mb2->m) 131 return(1); 132 return(0); 133 } 134 135 RB_GENERATE2(mbuf_rb_tree, mbtrack, rb_node, mbtrack_cmp, struct mbuf *, m); 136 137 struct mbuf_rb_tree mbuf_track_root; 138 static struct spinlock mbuf_track_spin = SPINLOCK_INITIALIZER(mbuf_track_spin, "mbuf_track_spin"); 139 140 static void 141 mbuftrack(struct mbuf *m) 142 { 143 struct mbtrack *mbt; 144 145 mbt = kmalloc(sizeof(*mbt), M_MTRACK, M_INTWAIT|M_ZERO); 146 spin_lock(&mbuf_track_spin); 147 mbt->m = m; 148 if (mbuf_rb_tree_RB_INSERT(&mbuf_track_root, mbt)) { 149 spin_unlock(&mbuf_track_spin); 150 panic("mbuftrack: mbuf %p already being tracked", m); 151 } 152 spin_unlock(&mbuf_track_spin); 153 } 154 155 static void 156 mbufuntrack(struct mbuf *m) 157 { 158 struct mbtrack *mbt; 159 160 spin_lock(&mbuf_track_spin); 161 mbt = mbuf_rb_tree_RB_LOOKUP(&mbuf_track_root, m); 162 if (mbt == NULL) { 163 spin_unlock(&mbuf_track_spin); 164 panic("mbufuntrack: mbuf %p was not tracked", m); 165 } else { 166 mbuf_rb_tree_RB_REMOVE(&mbuf_track_root, mbt); 167 spin_unlock(&mbuf_track_spin); 168 kfree(mbt, M_MTRACK); 169 } 170 } 171 172 void 173 mbuftrackid(struct mbuf *m, int trackid) 174 { 175 struct mbtrack *mbt; 176 struct mbuf *n; 177 178 spin_lock(&mbuf_track_spin); 179 while (m) { 180 n = m->m_nextpkt; 181 while (m) { 182 mbt = mbuf_rb_tree_RB_LOOKUP(&mbuf_track_root, m); 183 if (mbt == NULL) { 184 spin_unlock(&mbuf_track_spin); 185 panic("mbuftrackid: mbuf %p not tracked", m); 186 } 187 mbt->trackid = trackid; 188 m = m->m_next; 189 } 190 m = n; 191 } 192 spin_unlock(&mbuf_track_spin); 193 } 194 195 static int 196 mbuftrack_callback(struct mbtrack *mbt, void *arg) 197 { 198 struct sysctl_req *req = arg; 199 char buf[64]; 200 int error; 201 202 ksnprintf(buf, sizeof(buf), "mbuf %p track %d\n", mbt->m, mbt->trackid); 203 204 spin_unlock(&mbuf_track_spin); 205 error = SYSCTL_OUT(req, buf, strlen(buf)); 206 spin_lock(&mbuf_track_spin); 207 if (error) 208 return(-error); 209 return(0); 210 } 211 212 static int 213 mbuftrack_show(SYSCTL_HANDLER_ARGS) 214 { 215 int error; 216 217 spin_lock(&mbuf_track_spin); 218 error = mbuf_rb_tree_RB_SCAN(&mbuf_track_root, NULL, 219 mbuftrack_callback, req); 220 spin_unlock(&mbuf_track_spin); 221 return (-error); 222 } 223 SYSCTL_PROC(_kern_ipc, OID_AUTO, showmbufs, CTLFLAG_RD|CTLTYPE_STRING, 224 0, 0, mbuftrack_show, "A", "Show all in-use mbufs"); 225 226 #else 227 228 #define mbuftrack(m) 229 #define mbufuntrack(m) 230 231 #endif 232 233 static void mbinit(void *); 234 SYSINIT(mbuf, SI_BOOT2_MACHDEP, SI_ORDER_FIRST, mbinit, NULL); 235 236 struct mbtypes_stat { 237 u_long stats[MT_NTYPES]; 238 } __cachealign; 239 240 static struct mbtypes_stat mbtypes[SMP_MAXCPU]; 241 242 static struct mbstat mbstat[SMP_MAXCPU] __cachealign; 243 int max_linkhdr; 244 int max_protohdr; 245 int max_hdr; 246 int max_datalen; 247 int m_defragpackets; 248 int m_defragbytes; 249 int m_defraguseless; 250 int m_defragfailure; 251 #ifdef MBUF_STRESS_TEST 252 int m_defragrandomfailures; 253 #endif 254 255 struct objcache *mbuf_cache, *mbufphdr_cache; 256 struct objcache *mclmeta_cache, *mjclmeta_cache; 257 struct objcache *mbufcluster_cache, *mbufphdrcluster_cache; 258 struct objcache *mbufjcluster_cache, *mbufphdrjcluster_cache; 259 260 struct lock mbupdate_lk = LOCK_INITIALIZER("mbupdate", 0, LK_CANRECURSE); 261 262 int nmbclusters; 263 static int nmbjclusters; 264 int nmbufs; 265 266 static int mjclph_cachefrac; 267 static int mjcl_cachefrac; 268 static int mclph_cachefrac; 269 static int mcl_cachefrac; 270 271 SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RW, 272 &max_linkhdr, 0, "Max size of a link-level header"); 273 SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RW, 274 &max_protohdr, 0, "Max size of a protocol header"); 275 SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RW, &max_hdr, 0, 276 "Max size of link+protocol headers"); 277 SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RW, 278 &max_datalen, 0, "Max data payload size without headers"); 279 SYSCTL_INT(_kern_ipc, OID_AUTO, mbuf_wait, CTLFLAG_RW, 280 &mbuf_wait, 0, "Time in ticks to sleep after failed mbuf allocations"); 281 static int do_mbstat(SYSCTL_HANDLER_ARGS); 282 283 SYSCTL_PROC(_kern_ipc, KIPC_MBSTAT, mbstat, CTLTYPE_STRUCT|CTLFLAG_RD, 284 0, 0, do_mbstat, "S,mbstat", "mbuf usage statistics"); 285 286 static int do_mbtypes(SYSCTL_HANDLER_ARGS); 287 288 SYSCTL_PROC(_kern_ipc, OID_AUTO, mbtypes, CTLTYPE_ULONG|CTLFLAG_RD, 289 0, 0, do_mbtypes, "LU", ""); 290 291 static int 292 do_mbstat(SYSCTL_HANDLER_ARGS) 293 { 294 struct mbstat mbstat_total; 295 struct mbstat *mbstat_totalp; 296 int i; 297 298 bzero(&mbstat_total, sizeof(mbstat_total)); 299 mbstat_totalp = &mbstat_total; 300 301 for (i = 0; i < ncpus; i++) { 302 mbstat_total.m_mbufs += mbstat[i].m_mbufs; 303 mbstat_total.m_clusters += mbstat[i].m_clusters; 304 mbstat_total.m_jclusters += mbstat[i].m_jclusters; 305 mbstat_total.m_clfree += mbstat[i].m_clfree; 306 mbstat_total.m_drops += mbstat[i].m_drops; 307 mbstat_total.m_wait += mbstat[i].m_wait; 308 mbstat_total.m_drain += mbstat[i].m_drain; 309 mbstat_total.m_mcfail += mbstat[i].m_mcfail; 310 mbstat_total.m_mpfail += mbstat[i].m_mpfail; 311 312 } 313 /* 314 * The following fields are not cumulative fields so just 315 * get their values once. 316 */ 317 mbstat_total.m_msize = mbstat[0].m_msize; 318 mbstat_total.m_mclbytes = mbstat[0].m_mclbytes; 319 mbstat_total.m_minclsize = mbstat[0].m_minclsize; 320 mbstat_total.m_mlen = mbstat[0].m_mlen; 321 mbstat_total.m_mhlen = mbstat[0].m_mhlen; 322 323 return(sysctl_handle_opaque(oidp, mbstat_totalp, sizeof(mbstat_total), req)); 324 } 325 326 static int 327 do_mbtypes(SYSCTL_HANDLER_ARGS) 328 { 329 u_long totals[MT_NTYPES]; 330 int i, j; 331 332 for (i = 0; i < MT_NTYPES; i++) 333 totals[i] = 0; 334 335 for (i = 0; i < ncpus; i++) { 336 for (j = 0; j < MT_NTYPES; j++) 337 totals[j] += mbtypes[i].stats[j]; 338 } 339 340 return(sysctl_handle_opaque(oidp, totals, sizeof(totals), req)); 341 } 342 343 /* 344 * The variables may be set as boot-time tunables or live. Setting these 345 * values too low can deadlock your network. Network interfaces may also 346 * adjust nmbclusters and/or nmbjclusters to account for preloading the 347 * hardware rings. 348 */ 349 static int sysctl_nmbclusters(SYSCTL_HANDLER_ARGS); 350 static int sysctl_nmbjclusters(SYSCTL_HANDLER_ARGS); 351 static int sysctl_nmbufs(SYSCTL_HANDLER_ARGS); 352 SYSCTL_PROC(_kern_ipc, KIPC_NMBCLUSTERS, nmbclusters, CTLTYPE_INT | CTLFLAG_RW, 353 0, 0, sysctl_nmbclusters, "I", 354 "Maximum number of mbuf clusters available"); 355 SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjclusters, CTLTYPE_INT | CTLFLAG_RW, 356 0, 0, sysctl_nmbjclusters, "I", 357 "Maximum number of mbuf jclusters available"); 358 SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbufs, CTLTYPE_INT | CTLFLAG_RW, 359 0, 0, sysctl_nmbufs, "I", 360 "Maximum number of mbufs available"); 361 362 SYSCTL_INT(_kern_ipc, OID_AUTO, mjclph_cachefrac, CTLFLAG_RD, 363 &mjclph_cachefrac, 0, 364 "Fraction of cacheable mbuf jclusters w/ pkthdr"); 365 SYSCTL_INT(_kern_ipc, OID_AUTO, mjcl_cachefrac, CTLFLAG_RD, 366 &mjcl_cachefrac, 0, 367 "Fraction of cacheable mbuf jclusters"); 368 SYSCTL_INT(_kern_ipc, OID_AUTO, mclph_cachefrac, CTLFLAG_RD, 369 &mclph_cachefrac, 0, 370 "Fraction of cacheable mbuf clusters w/ pkthdr"); 371 SYSCTL_INT(_kern_ipc, OID_AUTO, mcl_cachefrac, CTLFLAG_RD, 372 &mcl_cachefrac, 0, "Fraction of cacheable mbuf clusters"); 373 374 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD, 375 &m_defragpackets, 0, "Number of defragment packets"); 376 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD, 377 &m_defragbytes, 0, "Number of defragment bytes"); 378 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD, 379 &m_defraguseless, 0, "Number of useless defragment mbuf chain operations"); 380 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD, 381 &m_defragfailure, 0, "Number of failed defragment mbuf chain operations"); 382 #ifdef MBUF_STRESS_TEST 383 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW, 384 &m_defragrandomfailures, 0, ""); 385 #endif 386 387 static MALLOC_DEFINE(M_MBUF, "mbuf", "mbuf"); 388 static MALLOC_DEFINE(M_MBUFCL, "mbufcl", "mbufcl"); 389 static MALLOC_DEFINE(M_MCLMETA, "mclmeta", "mclmeta"); 390 391 static void m_reclaim (void); 392 static void m_mclref(void *arg); 393 static void m_mclfree(void *arg); 394 static void m_mjclfree(void *arg); 395 396 static void mbupdatelimits(void); 397 398 /* 399 * Generally scale default mbufs to maxproc. 400 * 401 * NOTE: Default NMBUFS must take into account a possible DOS attack 402 * using fd passing on unix domain sockets. 403 */ 404 #ifndef NMBCLUSTERS 405 #define NMBCLUSTERS (512 + maxproc * 4) 406 #endif 407 #ifndef BASE_CACHEFRAC 408 #define BASE_CACHEFRAC 16 409 #endif 410 #ifndef MJCLPH_CACHEFRAC 411 #define MJCLPH_CACHEFRAC (BASE_CACHEFRAC * 2) 412 #endif 413 #ifndef MJCL_CACHEFRAC 414 #define MJCL_CACHEFRAC (BASE_CACHEFRAC * 2) 415 #endif 416 #ifndef MCLPH_CACHEFRAC 417 #define MCLPH_CACHEFRAC (BASE_CACHEFRAC * 2) 418 #endif 419 #ifndef MCL_CACHEFRAC 420 #define MCL_CACHEFRAC (BASE_CACHEFRAC * 2) 421 #endif 422 #ifndef NMBJCLUSTERS 423 #define NMBJCLUSTERS (NMBCLUSTERS / 4) 424 #endif 425 #ifndef NMBUFS 426 #define NMBUFS (nmbclusters / 2 + maxfiles) 427 #endif 428 429 #define NMBCLUSTERS_MIN (NMBCLUSTERS / 2) 430 #define NMBJCLUSTERS_MIN (NMBJCLUSTERS / 2) 431 #define NMBUFS_MIN (NMBUFS / 2) 432 433 /* 434 * Perform sanity checks of tunables declared above. 435 */ 436 static void 437 tunable_mbinit(void *dummy) 438 { 439 /* 440 * This has to be done before VM init. 441 */ 442 nmbclusters = NMBCLUSTERS; 443 TUNABLE_INT_FETCH("kern.ipc.nmbclusters", &nmbclusters); 444 mjclph_cachefrac = MJCLPH_CACHEFRAC; 445 TUNABLE_INT_FETCH("kern.ipc.mjclph_cachefrac", &mjclph_cachefrac); 446 mjcl_cachefrac = MJCL_CACHEFRAC; 447 TUNABLE_INT_FETCH("kern.ipc.mjcl_cachefrac", &mjcl_cachefrac); 448 mclph_cachefrac = MCLPH_CACHEFRAC; 449 TUNABLE_INT_FETCH("kern.ipc.mclph_cachefrac", &mclph_cachefrac); 450 mcl_cachefrac = MCL_CACHEFRAC; 451 TUNABLE_INT_FETCH("kern.ipc.mcl_cachefrac", &mcl_cachefrac); 452 453 /* 454 * WARNING! each mcl cache feeds two mbuf caches, so the minimum 455 * cachefrac is 2. For safety, use 3. 456 */ 457 if (mjclph_cachefrac < 3) 458 mjclph_cachefrac = 3; 459 if (mjcl_cachefrac < 3) 460 mjcl_cachefrac = 3; 461 if (mclph_cachefrac < 3) 462 mclph_cachefrac = 3; 463 if (mcl_cachefrac < 3) 464 mcl_cachefrac = 3; 465 466 nmbjclusters = NMBJCLUSTERS; 467 TUNABLE_INT_FETCH("kern.ipc.nmbjclusters", &nmbjclusters); 468 469 nmbufs = NMBUFS; 470 TUNABLE_INT_FETCH("kern.ipc.nmbufs", &nmbufs); 471 472 /* Sanity checks */ 473 if (nmbufs < nmbclusters * 2) 474 nmbufs = nmbclusters * 2; 475 } 476 SYSINIT(tunable_mbinit, SI_BOOT1_TUNABLES, SI_ORDER_ANY, 477 tunable_mbinit, NULL); 478 479 static void 480 mbinclimit(int *limit, int inc, int minlim) 481 { 482 int new_limit; 483 484 lockmgr(&mbupdate_lk, LK_EXCLUSIVE); 485 486 new_limit = *limit + inc; 487 if (new_limit < minlim) 488 new_limit = minlim; 489 490 if (*limit != new_limit) { 491 *limit = new_limit; 492 mbupdatelimits(); 493 } 494 495 lockmgr(&mbupdate_lk, LK_RELEASE); 496 } 497 498 static int 499 mbsetlimit(int *limit, int new_limit, int minlim) 500 { 501 if (new_limit < minlim) 502 return EINVAL; 503 504 lockmgr(&mbupdate_lk, LK_EXCLUSIVE); 505 mbinclimit(limit, new_limit - *limit, minlim); 506 lockmgr(&mbupdate_lk, LK_RELEASE); 507 return 0; 508 } 509 510 static int 511 sysctl_mblimit(SYSCTL_HANDLER_ARGS, int *limit, int minlim) 512 { 513 int error, value; 514 515 value = *limit; 516 error = sysctl_handle_int(oidp, &value, 0, req); 517 if (error || req->newptr == NULL) 518 return error; 519 520 return mbsetlimit(limit, value, minlim); 521 } 522 523 /* 524 * Sysctl support to update nmbclusters, nmbjclusters, and nmbufs. 525 */ 526 static int 527 sysctl_nmbclusters(SYSCTL_HANDLER_ARGS) 528 { 529 return sysctl_mblimit(oidp, arg1, arg2, req, &nmbclusters, 530 NMBCLUSTERS_MIN); 531 } 532 533 static int 534 sysctl_nmbjclusters(SYSCTL_HANDLER_ARGS) 535 { 536 return sysctl_mblimit(oidp, arg1, arg2, req, &nmbjclusters, 537 NMBJCLUSTERS_MIN); 538 } 539 540 static int 541 sysctl_nmbufs(SYSCTL_HANDLER_ARGS) 542 { 543 return sysctl_mblimit(oidp, arg1, arg2, req, &nmbufs, NMBUFS_MIN); 544 } 545 546 void 547 mcl_inclimit(int inc) 548 { 549 mbinclimit(&nmbclusters, inc, NMBCLUSTERS_MIN); 550 } 551 552 void 553 mjcl_inclimit(int inc) 554 { 555 mbinclimit(&nmbjclusters, inc, NMBJCLUSTERS_MIN); 556 } 557 558 void 559 mb_inclimit(int inc) 560 { 561 mbinclimit(&nmbufs, inc, NMBUFS_MIN); 562 } 563 564 /* "number of clusters of pages" */ 565 #define NCL_INIT 1 566 567 #define NMB_INIT 16 568 569 /* 570 * The mbuf object cache only guarantees that m_next and m_nextpkt are 571 * NULL and that m_data points to the beginning of the data area. In 572 * particular, m_len and m_pkthdr.len are uninitialized. It is the 573 * responsibility of the caller to initialize those fields before use. 574 */ 575 static __inline boolean_t 576 mbuf_ctor(void *obj, void *private, int ocflags) 577 { 578 struct mbuf *m = obj; 579 580 m->m_next = NULL; 581 m->m_nextpkt = NULL; 582 m->m_data = m->m_dat; 583 m->m_flags = 0; 584 585 return (TRUE); 586 } 587 588 /* 589 * Initialize the mbuf and the packet header fields. 590 */ 591 static boolean_t 592 mbufphdr_ctor(void *obj, void *private, int ocflags) 593 { 594 struct mbuf *m = obj; 595 596 m->m_next = NULL; 597 m->m_nextpkt = NULL; 598 m->m_data = m->m_pktdat; 599 m->m_flags = M_PKTHDR | M_PHCACHE; 600 601 m->m_pkthdr.rcvif = NULL; /* eliminate XXX JH */ 602 SLIST_INIT(&m->m_pkthdr.tags); 603 m->m_pkthdr.csum_flags = 0; /* eliminate XXX JH */ 604 m->m_pkthdr.fw_flags = 0; /* eliminate XXX JH */ 605 606 return (TRUE); 607 } 608 609 /* 610 * A mbcluster object consists of 2K (MCLBYTES) cluster and a refcount. 611 */ 612 static boolean_t 613 mclmeta_ctor(void *obj, void *private, int ocflags) 614 { 615 struct mbcluster *cl = obj; 616 void *buf; 617 618 if (ocflags & M_NOWAIT) 619 buf = kmalloc(MCLBYTES, M_MBUFCL, M_NOWAIT | M_ZERO); 620 else 621 buf = kmalloc(MCLBYTES, M_MBUFCL, M_INTWAIT | M_ZERO); 622 if (buf == NULL) 623 return (FALSE); 624 cl->mcl_refs = 0; 625 cl->mcl_data = buf; 626 return (TRUE); 627 } 628 629 static boolean_t 630 mjclmeta_ctor(void *obj, void *private, int ocflags) 631 { 632 struct mbcluster *cl = obj; 633 void *buf; 634 635 if (ocflags & M_NOWAIT) 636 buf = kmalloc(MJUMPAGESIZE, M_MBUFCL, M_NOWAIT | M_ZERO); 637 else 638 buf = kmalloc(MJUMPAGESIZE, M_MBUFCL, M_INTWAIT | M_ZERO); 639 if (buf == NULL) 640 return (FALSE); 641 cl->mcl_refs = 0; 642 cl->mcl_data = buf; 643 return (TRUE); 644 } 645 646 static void 647 mclmeta_dtor(void *obj, void *private) 648 { 649 struct mbcluster *mcl = obj; 650 651 KKASSERT(mcl->mcl_refs == 0); 652 kfree(mcl->mcl_data, M_MBUFCL); 653 } 654 655 static void 656 linkjcluster(struct mbuf *m, struct mbcluster *cl, uint size) 657 { 658 /* 659 * Add the cluster to the mbuf. The caller will detect that the 660 * mbuf now has an attached cluster. 661 */ 662 m->m_ext.ext_arg = cl; 663 m->m_ext.ext_buf = cl->mcl_data; 664 m->m_ext.ext_ref = m_mclref; 665 if (size != MCLBYTES) 666 m->m_ext.ext_free = m_mjclfree; 667 else 668 m->m_ext.ext_free = m_mclfree; 669 m->m_ext.ext_size = size; 670 atomic_add_int(&cl->mcl_refs, 1); 671 672 m->m_data = m->m_ext.ext_buf; 673 m->m_flags |= M_EXT | M_EXT_CLUSTER; 674 } 675 676 static void 677 linkcluster(struct mbuf *m, struct mbcluster *cl) 678 { 679 linkjcluster(m, cl, MCLBYTES); 680 } 681 682 static boolean_t 683 mbufphdrcluster_ctor(void *obj, void *private, int ocflags) 684 { 685 struct mbuf *m = obj; 686 struct mbcluster *cl; 687 688 mbufphdr_ctor(obj, private, ocflags); 689 cl = objcache_get(mclmeta_cache, ocflags); 690 if (cl == NULL) { 691 ++mbstat[mycpu->gd_cpuid].m_drops; 692 return (FALSE); 693 } 694 m->m_flags |= M_CLCACHE; 695 linkcluster(m, cl); 696 return (TRUE); 697 } 698 699 static boolean_t 700 mbufphdrjcluster_ctor(void *obj, void *private, int ocflags) 701 { 702 struct mbuf *m = obj; 703 struct mbcluster *cl; 704 705 mbufphdr_ctor(obj, private, ocflags); 706 cl = objcache_get(mjclmeta_cache, ocflags); 707 if (cl == NULL) { 708 ++mbstat[mycpu->gd_cpuid].m_drops; 709 return (FALSE); 710 } 711 m->m_flags |= M_CLCACHE; 712 linkjcluster(m, cl, MJUMPAGESIZE); 713 return (TRUE); 714 } 715 716 static boolean_t 717 mbufcluster_ctor(void *obj, void *private, int ocflags) 718 { 719 struct mbuf *m = obj; 720 struct mbcluster *cl; 721 722 mbuf_ctor(obj, private, ocflags); 723 cl = objcache_get(mclmeta_cache, ocflags); 724 if (cl == NULL) { 725 ++mbstat[mycpu->gd_cpuid].m_drops; 726 return (FALSE); 727 } 728 m->m_flags |= M_CLCACHE; 729 linkcluster(m, cl); 730 return (TRUE); 731 } 732 733 static boolean_t 734 mbufjcluster_ctor(void *obj, void *private, int ocflags) 735 { 736 struct mbuf *m = obj; 737 struct mbcluster *cl; 738 739 mbuf_ctor(obj, private, ocflags); 740 cl = objcache_get(mjclmeta_cache, ocflags); 741 if (cl == NULL) { 742 ++mbstat[mycpu->gd_cpuid].m_drops; 743 return (FALSE); 744 } 745 m->m_flags |= M_CLCACHE; 746 linkjcluster(m, cl, MJUMPAGESIZE); 747 return (TRUE); 748 } 749 750 /* 751 * Used for both the cluster and cluster PHDR caches. 752 * 753 * The mbuf may have lost its cluster due to sharing, deal 754 * with the situation by checking M_EXT. 755 */ 756 static void 757 mbufcluster_dtor(void *obj, void *private) 758 { 759 struct mbuf *m = obj; 760 struct mbcluster *mcl; 761 762 if (m->m_flags & M_EXT) { 763 KKASSERT((m->m_flags & M_EXT_CLUSTER) != 0); 764 mcl = m->m_ext.ext_arg; 765 KKASSERT(mcl->mcl_refs == 1); 766 mcl->mcl_refs = 0; 767 if (m->m_flags & M_EXT && m->m_ext.ext_size != MCLBYTES) 768 objcache_put(mjclmeta_cache, mcl); 769 else 770 objcache_put(mclmeta_cache, mcl); 771 } 772 } 773 774 struct objcache_malloc_args mbuf_malloc_args = { MSIZE, M_MBUF }; 775 struct objcache_malloc_args mclmeta_malloc_args = 776 { sizeof(struct mbcluster), M_MCLMETA }; 777 778 /* ARGSUSED*/ 779 static void 780 mbinit(void *dummy) 781 { 782 int mb_limit, cl_limit, ncl_limit, jcl_limit; 783 int limit; 784 int i; 785 786 /* 787 * Initialize statistics 788 */ 789 for (i = 0; i < ncpus; i++) { 790 mbstat[i].m_msize = MSIZE; 791 mbstat[i].m_mclbytes = MCLBYTES; 792 mbstat[i].m_mjumpagesize = MJUMPAGESIZE; 793 mbstat[i].m_minclsize = MINCLSIZE; 794 mbstat[i].m_mlen = MLEN; 795 mbstat[i].m_mhlen = MHLEN; 796 } 797 798 /* 799 * Create object caches and save cluster limits, which will 800 * be used to adjust backing kmalloc pools' limit later. 801 */ 802 803 mb_limit = cl_limit = 0; 804 805 limit = nmbufs; 806 mbuf_cache = objcache_create("mbuf", 807 limit, nmbufs / BASE_CACHEFRAC, 808 mbuf_ctor, NULL, NULL, 809 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args); 810 mb_limit += limit; 811 812 limit = nmbufs; 813 mbufphdr_cache = objcache_create("mbuf pkt hdr", 814 limit, nmbufs / BASE_CACHEFRAC, 815 mbufphdr_ctor, NULL, NULL, 816 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args); 817 mb_limit += limit; 818 819 ncl_limit = nmbclusters; 820 mclmeta_cache = objcache_create("cluster mbuf", 821 ncl_limit, nmbclusters / BASE_CACHEFRAC, 822 mclmeta_ctor, mclmeta_dtor, NULL, 823 objcache_malloc_alloc, objcache_malloc_free, &mclmeta_malloc_args); 824 cl_limit += ncl_limit; 825 826 jcl_limit = nmbjclusters; 827 mjclmeta_cache = objcache_create("jcluster mbuf", 828 jcl_limit, nmbjclusters / BASE_CACHEFRAC, 829 mjclmeta_ctor, mclmeta_dtor, NULL, 830 objcache_malloc_alloc, objcache_malloc_free, &mclmeta_malloc_args); 831 cl_limit += jcl_limit; 832 833 limit = nmbclusters; 834 mbufcluster_cache = objcache_create("mbuf + cluster", 835 limit, nmbclusters / mcl_cachefrac, 836 mbufcluster_ctor, mbufcluster_dtor, NULL, 837 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args); 838 mb_limit += limit; 839 840 limit = nmbclusters; 841 mbufphdrcluster_cache = objcache_create("mbuf pkt hdr + cluster", 842 limit, nmbclusters / mclph_cachefrac, 843 mbufphdrcluster_ctor, mbufcluster_dtor, NULL, 844 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args); 845 mb_limit += limit; 846 847 limit = nmbjclusters; 848 mbufjcluster_cache = objcache_create("mbuf + jcluster", 849 limit, nmbjclusters / mjcl_cachefrac, 850 mbufjcluster_ctor, mbufcluster_dtor, NULL, 851 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args); 852 mb_limit += limit; 853 854 limit = nmbjclusters; 855 mbufphdrjcluster_cache = objcache_create("mbuf pkt hdr + jcluster", 856 limit, nmbjclusters / mjclph_cachefrac, 857 mbufphdrjcluster_ctor, mbufcluster_dtor, NULL, 858 objcache_malloc_alloc, objcache_malloc_free, &mbuf_malloc_args); 859 mb_limit += limit; 860 861 /* 862 * Adjust backing kmalloc pools' limit 863 * 864 * NOTE: We raise the limit by another 1/8 to take the effect 865 * of loosememuse into account. 866 */ 867 cl_limit += cl_limit / 8; 868 kmalloc_raise_limit(mclmeta_malloc_args.mtype, 869 mclmeta_malloc_args.objsize * (size_t)cl_limit); 870 kmalloc_raise_limit(M_MBUFCL, 871 (MCLBYTES * (size_t)ncl_limit) + 872 (MJUMPAGESIZE * (size_t)jcl_limit)); 873 874 mb_limit += mb_limit / 8; 875 kmalloc_raise_limit(mbuf_malloc_args.mtype, 876 mbuf_malloc_args.objsize * (size_t)mb_limit); 877 } 878 879 /* 880 * Adjust mbuf limits after changes have been made 881 * 882 * Caller must hold mbupdate_lk 883 */ 884 static void 885 mbupdatelimits(void) 886 { 887 int mb_limit, cl_limit, ncl_limit, jcl_limit; 888 int limit; 889 890 KASSERT(lockstatus(&mbupdate_lk, curthread) != 0, 891 ("mbupdate_lk is not held")); 892 893 /* 894 * Figure out adjustments to object caches after nmbufs, nmbclusters, 895 * or nmbjclusters has been modified. 896 */ 897 mb_limit = cl_limit = 0; 898 899 limit = nmbufs; 900 objcache_set_cluster_limit(mbuf_cache, limit); 901 mb_limit += limit; 902 903 limit = nmbufs; 904 objcache_set_cluster_limit(mbufphdr_cache, limit); 905 mb_limit += limit; 906 907 ncl_limit = nmbclusters; 908 objcache_set_cluster_limit(mclmeta_cache, ncl_limit); 909 cl_limit += ncl_limit; 910 911 jcl_limit = nmbjclusters; 912 objcache_set_cluster_limit(mjclmeta_cache, jcl_limit); 913 cl_limit += jcl_limit; 914 915 limit = nmbclusters; 916 objcache_set_cluster_limit(mbufcluster_cache, limit); 917 mb_limit += limit; 918 919 limit = nmbclusters; 920 objcache_set_cluster_limit(mbufphdrcluster_cache, limit); 921 mb_limit += limit; 922 923 limit = nmbjclusters; 924 objcache_set_cluster_limit(mbufjcluster_cache, limit); 925 mb_limit += limit; 926 927 limit = nmbjclusters; 928 objcache_set_cluster_limit(mbufphdrjcluster_cache, limit); 929 mb_limit += limit; 930 931 /* 932 * Adjust backing kmalloc pools' limit 933 * 934 * NOTE: We raise the limit by another 1/8 to take the effect 935 * of loosememuse into account. 936 */ 937 cl_limit += cl_limit / 8; 938 kmalloc_raise_limit(mclmeta_malloc_args.mtype, 939 mclmeta_malloc_args.objsize * (size_t)cl_limit); 940 kmalloc_raise_limit(M_MBUFCL, 941 (MCLBYTES * (size_t)ncl_limit) + 942 (MJUMPAGESIZE * (size_t)jcl_limit)); 943 mb_limit += mb_limit / 8; 944 kmalloc_raise_limit(mbuf_malloc_args.mtype, 945 mbuf_malloc_args.objsize * (size_t)mb_limit); 946 } 947 948 /* 949 * Return the number of references to this mbuf's data. 0 is returned 950 * if the mbuf is not M_EXT, a reference count is returned if it is 951 * M_EXT | M_EXT_CLUSTER, and 99 is returned if it is a special M_EXT. 952 */ 953 int 954 m_sharecount(struct mbuf *m) 955 { 956 switch (m->m_flags & (M_EXT | M_EXT_CLUSTER)) { 957 case 0: 958 return (0); 959 case M_EXT: 960 return (99); 961 case M_EXT | M_EXT_CLUSTER: 962 return (((struct mbcluster *)m->m_ext.ext_arg)->mcl_refs); 963 } 964 /* NOTREACHED */ 965 return (0); /* to shut up compiler */ 966 } 967 968 /* 969 * change mbuf to new type 970 */ 971 void 972 m_chtype(struct mbuf *m, int type) 973 { 974 struct globaldata *gd = mycpu; 975 976 ++mbtypes[gd->gd_cpuid].stats[type]; 977 --mbtypes[gd->gd_cpuid].stats[m->m_type]; 978 m->m_type = type; 979 } 980 981 static void 982 m_reclaim(void) 983 { 984 struct domain *dp; 985 struct protosw *pr; 986 987 kprintf("Debug: m_reclaim() called\n"); 988 989 SLIST_FOREACH(dp, &domains, dom_next) { 990 for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) { 991 if (pr->pr_drain) 992 (*pr->pr_drain)(); 993 } 994 } 995 ++mbstat[mycpu->gd_cpuid].m_drain; 996 } 997 998 static __inline void 999 updatestats(struct mbuf *m, int type) 1000 { 1001 struct globaldata *gd = mycpu; 1002 1003 m->m_type = type; 1004 mbuftrack(m); 1005 #ifdef MBUF_DEBUG 1006 KASSERT(m->m_next == NULL, ("mbuf %p: bad m_next in get", m)); 1007 KASSERT(m->m_nextpkt == NULL, ("mbuf %p: bad m_nextpkt in get", m)); 1008 #endif 1009 1010 ++mbtypes[gd->gd_cpuid].stats[type]; 1011 ++mbstat[gd->gd_cpuid].m_mbufs; 1012 1013 } 1014 1015 /* 1016 * Allocate an mbuf. 1017 */ 1018 struct mbuf * 1019 m_get(int how, int type) 1020 { 1021 struct mbuf *m; 1022 int ntries = 0; 1023 int ocf = MB_OCFLAG(how); 1024 1025 retryonce: 1026 1027 m = objcache_get(mbuf_cache, ocf); 1028 1029 if (m == NULL) { 1030 if ((ocf & M_WAITOK) && ntries++ == 0) { 1031 struct objcache *reclaimlist[] = { 1032 mbufphdr_cache, 1033 mbufcluster_cache, 1034 mbufphdrcluster_cache, 1035 mbufjcluster_cache, 1036 mbufphdrjcluster_cache 1037 }; 1038 const int nreclaims = NELEM(reclaimlist); 1039 1040 if (!objcache_reclaimlist(reclaimlist, nreclaims, ocf)) 1041 m_reclaim(); 1042 goto retryonce; 1043 } 1044 ++mbstat[mycpu->gd_cpuid].m_drops; 1045 return (NULL); 1046 } 1047 #ifdef MBUF_DEBUG 1048 KASSERT(m->m_data == m->m_dat, ("mbuf %p: bad m_data in get", m)); 1049 #endif 1050 m->m_len = 0; 1051 1052 updatestats(m, type); 1053 return (m); 1054 } 1055 1056 struct mbuf * 1057 m_gethdr(int how, int type) 1058 { 1059 struct mbuf *m; 1060 int ocf = MB_OCFLAG(how); 1061 int ntries = 0; 1062 1063 retryonce: 1064 1065 m = objcache_get(mbufphdr_cache, ocf); 1066 1067 if (m == NULL) { 1068 if ((ocf & M_WAITOK) && ntries++ == 0) { 1069 struct objcache *reclaimlist[] = { 1070 mbuf_cache, 1071 mbufcluster_cache, mbufphdrcluster_cache, 1072 mbufjcluster_cache, mbufphdrjcluster_cache 1073 }; 1074 const int nreclaims = NELEM(reclaimlist); 1075 1076 if (!objcache_reclaimlist(reclaimlist, nreclaims, ocf)) 1077 m_reclaim(); 1078 goto retryonce; 1079 } 1080 ++mbstat[mycpu->gd_cpuid].m_drops; 1081 return (NULL); 1082 } 1083 #ifdef MBUF_DEBUG 1084 KASSERT(m->m_data == m->m_pktdat, ("mbuf %p: bad m_data in get", m)); 1085 #endif 1086 m->m_len = 0; 1087 m->m_pkthdr.len = 0; 1088 1089 updatestats(m, type); 1090 return (m); 1091 } 1092 1093 /* 1094 * Get a mbuf (not a mbuf cluster!) and zero it. 1095 * Deprecated. 1096 */ 1097 struct mbuf * 1098 m_getclr(int how, int type) 1099 { 1100 struct mbuf *m; 1101 1102 m = m_get(how, type); 1103 if (m != NULL) 1104 bzero(m->m_data, MLEN); 1105 return (m); 1106 } 1107 1108 static struct mbuf * 1109 m_getcl_cache(int how, short type, int flags, struct objcache *mbclc, 1110 struct objcache *mbphclc, u_long *cl_stats) 1111 { 1112 struct mbuf *m = NULL; 1113 int ocflags = MB_OCFLAG(how); 1114 int ntries = 0; 1115 1116 retryonce: 1117 1118 if (flags & M_PKTHDR) 1119 m = objcache_get(mbphclc, ocflags); 1120 else 1121 m = objcache_get(mbclc, ocflags); 1122 1123 if (m == NULL) { 1124 if ((ocflags & M_WAITOK) && ntries++ == 0) { 1125 struct objcache *reclaimlist[1]; 1126 1127 if (flags & M_PKTHDR) 1128 reclaimlist[0] = mbclc; 1129 else 1130 reclaimlist[0] = mbphclc; 1131 if (!objcache_reclaimlist(reclaimlist, 1, ocflags)) 1132 m_reclaim(); 1133 goto retryonce; 1134 } 1135 ++mbstat[mycpu->gd_cpuid].m_drops; 1136 return (NULL); 1137 } 1138 1139 #ifdef MBUF_DEBUG 1140 KASSERT(m->m_data == m->m_ext.ext_buf, 1141 ("mbuf %p: bad m_data in get", m)); 1142 #endif 1143 m->m_type = type; 1144 m->m_len = 0; 1145 m->m_pkthdr.len = 0; /* just do it unconditonally */ 1146 1147 mbuftrack(m); 1148 1149 ++mbtypes[mycpu->gd_cpuid].stats[type]; 1150 ++(*cl_stats); 1151 return (m); 1152 } 1153 1154 struct mbuf * 1155 m_getjcl(int how, short type, int flags, size_t size) 1156 { 1157 struct objcache *mbclc, *mbphclc; 1158 u_long *cl_stats; 1159 1160 switch (size) { 1161 case MCLBYTES: 1162 mbclc = mbufcluster_cache; 1163 mbphclc = mbufphdrcluster_cache; 1164 cl_stats = &mbstat[mycpu->gd_cpuid].m_clusters; 1165 break; 1166 1167 default: 1168 mbclc = mbufjcluster_cache; 1169 mbphclc = mbufphdrjcluster_cache; 1170 cl_stats = &mbstat[mycpu->gd_cpuid].m_jclusters; 1171 break; 1172 } 1173 return m_getcl_cache(how, type, flags, mbclc, mbphclc, cl_stats); 1174 } 1175 1176 /* 1177 * Returns an mbuf with an attached cluster. 1178 * Because many network drivers use this kind of buffers a lot, it is 1179 * convenient to keep a small pool of free buffers of this kind. 1180 * Even a small size such as 10 gives about 10% improvement in the 1181 * forwarding rate in a bridge or router. 1182 */ 1183 struct mbuf * 1184 m_getcl(int how, short type, int flags) 1185 { 1186 return m_getcl_cache(how, type, flags, 1187 mbufcluster_cache, mbufphdrcluster_cache, 1188 &mbstat[mycpu->gd_cpuid].m_clusters); 1189 } 1190 1191 /* 1192 * Allocate chain of requested length. 1193 */ 1194 struct mbuf * 1195 m_getc(int len, int how, int type) 1196 { 1197 struct mbuf *n, *nfirst = NULL, **ntail = &nfirst; 1198 int nsize; 1199 1200 while (len > 0) { 1201 n = m_getl(len, how, type, 0, &nsize); 1202 if (n == NULL) 1203 goto failed; 1204 n->m_len = 0; 1205 *ntail = n; 1206 ntail = &n->m_next; 1207 len -= nsize; 1208 } 1209 return (nfirst); 1210 1211 failed: 1212 m_freem(nfirst); 1213 return (NULL); 1214 } 1215 1216 /* 1217 * Allocate len-worth of mbufs and/or mbuf clusters (whatever fits best) 1218 * and return a pointer to the head of the allocated chain. If m0 is 1219 * non-null, then we assume that it is a single mbuf or an mbuf chain to 1220 * which we want len bytes worth of mbufs and/or clusters attached, and so 1221 * if we succeed in allocating it, we will just return a pointer to m0. 1222 * 1223 * If we happen to fail at any point during the allocation, we will free 1224 * up everything we have already allocated and return NULL. 1225 * 1226 * Deprecated. Use m_getc() and m_cat() instead. 1227 */ 1228 struct mbuf * 1229 m_getm(struct mbuf *m0, int len, int type, int how) 1230 { 1231 struct mbuf *nfirst; 1232 1233 nfirst = m_getc(len, how, type); 1234 1235 if (m0 != NULL) { 1236 m_last(m0)->m_next = nfirst; 1237 return (m0); 1238 } 1239 1240 return (nfirst); 1241 } 1242 1243 /* 1244 * Adds a cluster to a normal mbuf, M_EXT is set on success. 1245 * Deprecated. Use m_getcl() instead. 1246 */ 1247 void 1248 m_mclget(struct mbuf *m, int how) 1249 { 1250 struct mbcluster *mcl; 1251 1252 KKASSERT((m->m_flags & M_EXT) == 0); 1253 mcl = objcache_get(mclmeta_cache, MB_OCFLAG(how)); 1254 if (mcl != NULL) { 1255 linkcluster(m, mcl); 1256 ++mbstat[mycpu->gd_cpuid].m_clusters; 1257 } else { 1258 ++mbstat[mycpu->gd_cpuid].m_drops; 1259 } 1260 } 1261 1262 /* 1263 * Updates to mbcluster must be MPSAFE. Only an entity which already has 1264 * a reference to the cluster can ref it, so we are in no danger of 1265 * racing an add with a subtract. But the operation must still be atomic 1266 * since multiple entities may have a reference on the cluster. 1267 * 1268 * m_mclfree() is almost the same but it must contend with two entities 1269 * freeing the cluster at the same time. 1270 */ 1271 static void 1272 m_mclref(void *arg) 1273 { 1274 struct mbcluster *mcl = arg; 1275 1276 atomic_add_int(&mcl->mcl_refs, 1); 1277 } 1278 1279 /* 1280 * When dereferencing a cluster we have to deal with a N->0 race, where 1281 * N entities free their references simultaniously. To do this we use 1282 * atomic_fetchadd_int(). 1283 */ 1284 static void 1285 m_mclfree(void *arg) 1286 { 1287 struct mbcluster *mcl = arg; 1288 1289 if (atomic_fetchadd_int(&mcl->mcl_refs, -1) == 1) { 1290 --mbstat[mycpu->gd_cpuid].m_clusters; 1291 objcache_put(mclmeta_cache, mcl); 1292 } 1293 } 1294 1295 static void 1296 m_mjclfree(void *arg) 1297 { 1298 struct mbcluster *mcl = arg; 1299 1300 if (atomic_fetchadd_int(&mcl->mcl_refs, -1) == 1) { 1301 --mbstat[mycpu->gd_cpuid].m_jclusters; 1302 objcache_put(mjclmeta_cache, mcl); 1303 } 1304 } 1305 1306 /* 1307 * Free a single mbuf and any associated external storage. The successor, 1308 * if any, is returned. 1309 * 1310 * We do need to check non-first mbuf for m_aux, since some of existing 1311 * code does not call M_PREPEND properly. 1312 * (example: call to bpf_mtap from drivers) 1313 */ 1314 1315 #ifdef MBUF_DEBUG 1316 1317 struct mbuf * 1318 _m_free(struct mbuf *m, const char *func) 1319 1320 #else 1321 1322 struct mbuf * 1323 m_free(struct mbuf *m) 1324 1325 #endif 1326 { 1327 struct mbuf *n; 1328 struct globaldata *gd = mycpu; 1329 1330 KASSERT(m->m_type != MT_FREE, ("freeing free mbuf %p", m)); 1331 KASSERT(M_TRAILINGSPACE(m) >= 0, ("overflowed mbuf %p", m)); 1332 --mbtypes[gd->gd_cpuid].stats[m->m_type]; 1333 1334 n = m->m_next; 1335 1336 /* 1337 * Make sure the mbuf is in constructed state before returning it 1338 * to the objcache. 1339 */ 1340 m->m_next = NULL; 1341 mbufuntrack(m); 1342 #ifdef MBUF_DEBUG 1343 m->m_hdr.mh_lastfunc = func; 1344 #endif 1345 #ifdef notyet 1346 KKASSERT(m->m_nextpkt == NULL); 1347 #else 1348 if (m->m_nextpkt != NULL) { 1349 static int afewtimes = 10; 1350 1351 if (afewtimes-- > 0) { 1352 kprintf("mfree: m->m_nextpkt != NULL\n"); 1353 print_backtrace(-1); 1354 } 1355 m->m_nextpkt = NULL; 1356 } 1357 #endif 1358 if (m->m_flags & M_PKTHDR) { 1359 m_tag_delete_chain(m); /* eliminate XXX JH */ 1360 } 1361 1362 m->m_flags &= (M_EXT | M_EXT_CLUSTER | M_CLCACHE | M_PHCACHE); 1363 1364 /* 1365 * Clean the M_PKTHDR state so we can return the mbuf to its original 1366 * cache. This is based on the PHCACHE flag which tells us whether 1367 * the mbuf was originally allocated out of a packet-header cache 1368 * or a non-packet-header cache. 1369 */ 1370 if (m->m_flags & M_PHCACHE) { 1371 m->m_flags |= M_PKTHDR; 1372 m->m_pkthdr.rcvif = NULL; /* eliminate XXX JH */ 1373 m->m_pkthdr.csum_flags = 0; /* eliminate XXX JH */ 1374 m->m_pkthdr.fw_flags = 0; /* eliminate XXX JH */ 1375 SLIST_INIT(&m->m_pkthdr.tags); 1376 } 1377 1378 /* 1379 * Handle remaining flags combinations. M_CLCACHE tells us whether 1380 * the mbuf was originally allocated from a cluster cache or not, 1381 * and is totally separate from whether the mbuf is currently 1382 * associated with a cluster. 1383 */ 1384 switch(m->m_flags & (M_CLCACHE | M_EXT | M_EXT_CLUSTER)) { 1385 case M_CLCACHE | M_EXT | M_EXT_CLUSTER: 1386 /* 1387 * mbuf+cluster cache case. The mbuf was allocated from the 1388 * combined mbuf_cluster cache and can be returned to the 1389 * cache if the cluster hasn't been shared. 1390 */ 1391 if (m_sharecount(m) == 1) { 1392 /* 1393 * The cluster has not been shared, we can just 1394 * reset the data pointer and return the mbuf 1395 * to the cluster cache. Note that the reference 1396 * count is left intact (it is still associated with 1397 * an mbuf). 1398 */ 1399 m->m_data = m->m_ext.ext_buf; 1400 if (m->m_flags & M_EXT && m->m_ext.ext_size != MCLBYTES) { 1401 if (m->m_flags & M_PHCACHE) 1402 objcache_put(mbufphdrjcluster_cache, m); 1403 else 1404 objcache_put(mbufjcluster_cache, m); 1405 --mbstat[mycpu->gd_cpuid].m_jclusters; 1406 } else { 1407 if (m->m_flags & M_PHCACHE) 1408 objcache_put(mbufphdrcluster_cache, m); 1409 else 1410 objcache_put(mbufcluster_cache, m); 1411 --mbstat[mycpu->gd_cpuid].m_clusters; 1412 } 1413 } else { 1414 /* 1415 * Hell. Someone else has a ref on this cluster, 1416 * we have to disconnect it which means we can't 1417 * put it back into the mbufcluster_cache, we 1418 * have to destroy the mbuf. 1419 * 1420 * Other mbuf references to the cluster will typically 1421 * be M_EXT | M_EXT_CLUSTER but without M_CLCACHE. 1422 * 1423 * XXX we could try to connect another cluster to 1424 * it. 1425 */ 1426 m->m_ext.ext_free(m->m_ext.ext_arg); 1427 m->m_flags &= ~(M_EXT | M_EXT_CLUSTER); 1428 if (m->m_ext.ext_size == MCLBYTES) { 1429 if (m->m_flags & M_PHCACHE) 1430 objcache_dtor(mbufphdrcluster_cache, m); 1431 else 1432 objcache_dtor(mbufcluster_cache, m); 1433 } else { 1434 if (m->m_flags & M_PHCACHE) 1435 objcache_dtor(mbufphdrjcluster_cache, m); 1436 else 1437 objcache_dtor(mbufjcluster_cache, m); 1438 } 1439 } 1440 break; 1441 case M_EXT | M_EXT_CLUSTER: 1442 case M_EXT: 1443 /* 1444 * Normal cluster association case, disconnect the cluster from 1445 * the mbuf. The cluster may or may not be custom. 1446 */ 1447 m->m_ext.ext_free(m->m_ext.ext_arg); 1448 m->m_flags &= ~(M_EXT | M_EXT_CLUSTER); 1449 /* fall through */ 1450 case 0: 1451 /* 1452 * return the mbuf to the mbuf cache. 1453 */ 1454 if (m->m_flags & M_PHCACHE) { 1455 m->m_data = m->m_pktdat; 1456 objcache_put(mbufphdr_cache, m); 1457 } else { 1458 m->m_data = m->m_dat; 1459 objcache_put(mbuf_cache, m); 1460 } 1461 --mbstat[mycpu->gd_cpuid].m_mbufs; 1462 break; 1463 default: 1464 if (!panicstr) 1465 panic("bad mbuf flags %p %08x", m, m->m_flags); 1466 break; 1467 } 1468 return (n); 1469 } 1470 1471 #ifdef MBUF_DEBUG 1472 1473 void 1474 _m_freem(struct mbuf *m, const char *func) 1475 { 1476 while (m) 1477 m = _m_free(m, func); 1478 } 1479 1480 #else 1481 1482 void 1483 m_freem(struct mbuf *m) 1484 { 1485 while (m) 1486 m = m_free(m); 1487 } 1488 1489 #endif 1490 1491 void 1492 m_extadd(struct mbuf *m, caddr_t buf, u_int size, void (*reff)(void *), 1493 void (*freef)(void *), void *arg) 1494 { 1495 m->m_ext.ext_arg = arg; 1496 m->m_ext.ext_buf = buf; 1497 m->m_ext.ext_ref = reff; 1498 m->m_ext.ext_free = freef; 1499 m->m_ext.ext_size = size; 1500 reff(arg); 1501 m->m_data = buf; 1502 m->m_flags |= M_EXT; 1503 } 1504 1505 /* 1506 * mbuf utility routines 1507 */ 1508 1509 /* 1510 * Lesser-used path for M_PREPEND: allocate new mbuf to prepend to chain and 1511 * copy junk along. 1512 */ 1513 struct mbuf * 1514 m_prepend(struct mbuf *m, int len, int how) 1515 { 1516 struct mbuf *mn; 1517 1518 if (m->m_flags & M_PKTHDR) 1519 mn = m_gethdr(how, m->m_type); 1520 else 1521 mn = m_get(how, m->m_type); 1522 if (mn == NULL) { 1523 m_freem(m); 1524 return (NULL); 1525 } 1526 if (m->m_flags & M_PKTHDR) 1527 M_MOVE_PKTHDR(mn, m); 1528 mn->m_next = m; 1529 m = mn; 1530 if (len < MHLEN) 1531 MH_ALIGN(m, len); 1532 m->m_len = len; 1533 return (m); 1534 } 1535 1536 /* 1537 * Make a copy of an mbuf chain starting "off0" bytes from the beginning, 1538 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf. 1539 * The wait parameter is a choice of M_WAITOK/M_NOWAIT from caller. 1540 * Note that the copy is read-only, because clusters are not copied, 1541 * only their reference counts are incremented. 1542 */ 1543 struct mbuf * 1544 m_copym(const struct mbuf *m, int off0, int len, int wait) 1545 { 1546 struct mbuf *n, **np; 1547 int off = off0; 1548 struct mbuf *top; 1549 int copyhdr = 0; 1550 1551 KASSERT(off >= 0, ("m_copym, negative off %d", off)); 1552 KASSERT(len >= 0, ("m_copym, negative len %d", len)); 1553 if (off == 0 && (m->m_flags & M_PKTHDR)) 1554 copyhdr = 1; 1555 while (off > 0) { 1556 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain")); 1557 if (off < m->m_len) 1558 break; 1559 off -= m->m_len; 1560 m = m->m_next; 1561 } 1562 np = ⊤ 1563 top = NULL; 1564 while (len > 0) { 1565 if (m == NULL) { 1566 KASSERT(len == M_COPYALL, 1567 ("m_copym, length > size of mbuf chain")); 1568 break; 1569 } 1570 /* 1571 * Because we are sharing any cluster attachment below, 1572 * be sure to get an mbuf that does not have a cluster 1573 * associated with it. 1574 */ 1575 if (copyhdr) 1576 n = m_gethdr(wait, m->m_type); 1577 else 1578 n = m_get(wait, m->m_type); 1579 *np = n; 1580 if (n == NULL) 1581 goto nospace; 1582 if (copyhdr) { 1583 if (!m_dup_pkthdr(n, m, wait)) 1584 goto nospace; 1585 if (len == M_COPYALL) 1586 n->m_pkthdr.len -= off0; 1587 else 1588 n->m_pkthdr.len = len; 1589 copyhdr = 0; 1590 } 1591 n->m_len = min(len, m->m_len - off); 1592 if (m->m_flags & M_EXT) { 1593 KKASSERT((n->m_flags & M_EXT) == 0); 1594 n->m_data = m->m_data + off; 1595 m->m_ext.ext_ref(m->m_ext.ext_arg); 1596 n->m_ext = m->m_ext; 1597 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER); 1598 } else { 1599 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t), 1600 (unsigned)n->m_len); 1601 } 1602 if (len != M_COPYALL) 1603 len -= n->m_len; 1604 off = 0; 1605 m = m->m_next; 1606 np = &n->m_next; 1607 } 1608 if (top == NULL) 1609 ++mbstat[mycpu->gd_cpuid].m_mcfail; 1610 return (top); 1611 nospace: 1612 m_freem(top); 1613 ++mbstat[mycpu->gd_cpuid].m_mcfail; 1614 return (NULL); 1615 } 1616 1617 /* 1618 * Copy an entire packet, including header (which must be present). 1619 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'. 1620 * Note that the copy is read-only, because clusters are not copied, 1621 * only their reference counts are incremented. 1622 * Preserve alignment of the first mbuf so if the creator has left 1623 * some room at the beginning (e.g. for inserting protocol headers) 1624 * the copies also have the room available. 1625 */ 1626 struct mbuf * 1627 m_copypacket(struct mbuf *m, int how) 1628 { 1629 struct mbuf *top, *n, *o; 1630 1631 n = m_gethdr(how, m->m_type); 1632 top = n; 1633 if (!n) 1634 goto nospace; 1635 1636 if (!m_dup_pkthdr(n, m, how)) 1637 goto nospace; 1638 n->m_len = m->m_len; 1639 if (m->m_flags & M_EXT) { 1640 KKASSERT((n->m_flags & M_EXT) == 0); 1641 n->m_data = m->m_data; 1642 m->m_ext.ext_ref(m->m_ext.ext_arg); 1643 n->m_ext = m->m_ext; 1644 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER); 1645 } else { 1646 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat ); 1647 bcopy(mtod(m, char *), mtod(n, char *), n->m_len); 1648 } 1649 1650 m = m->m_next; 1651 while (m) { 1652 o = m_get(how, m->m_type); 1653 if (!o) 1654 goto nospace; 1655 1656 n->m_next = o; 1657 n = n->m_next; 1658 1659 n->m_len = m->m_len; 1660 if (m->m_flags & M_EXT) { 1661 KKASSERT((n->m_flags & M_EXT) == 0); 1662 n->m_data = m->m_data; 1663 m->m_ext.ext_ref(m->m_ext.ext_arg); 1664 n->m_ext = m->m_ext; 1665 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER); 1666 } else { 1667 bcopy(mtod(m, char *), mtod(n, char *), n->m_len); 1668 } 1669 1670 m = m->m_next; 1671 } 1672 return top; 1673 nospace: 1674 m_freem(top); 1675 ++mbstat[mycpu->gd_cpuid].m_mcfail; 1676 return (NULL); 1677 } 1678 1679 /* 1680 * Copy data from an mbuf chain starting "off" bytes from the beginning, 1681 * continuing for "len" bytes, into the indicated buffer. 1682 */ 1683 void 1684 m_copydata(const struct mbuf *m, int off, int len, caddr_t cp) 1685 { 1686 unsigned count; 1687 1688 KASSERT(off >= 0, ("m_copydata, negative off %d", off)); 1689 KASSERT(len >= 0, ("m_copydata, negative len %d", len)); 1690 while (off > 0) { 1691 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain")); 1692 if (off < m->m_len) 1693 break; 1694 off -= m->m_len; 1695 m = m->m_next; 1696 } 1697 while (len > 0) { 1698 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain")); 1699 count = min(m->m_len - off, len); 1700 bcopy(mtod(m, caddr_t) + off, cp, count); 1701 len -= count; 1702 cp += count; 1703 off = 0; 1704 m = m->m_next; 1705 } 1706 } 1707 1708 /* 1709 * Copy a packet header mbuf chain into a completely new chain, including 1710 * copying any mbuf clusters. Use this instead of m_copypacket() when 1711 * you need a writable copy of an mbuf chain. 1712 */ 1713 struct mbuf * 1714 m_dup(struct mbuf *m, int how) 1715 { 1716 struct mbuf **p, *top = NULL; 1717 int remain, moff, nsize; 1718 1719 /* Sanity check */ 1720 if (m == NULL) 1721 return (NULL); 1722 KASSERT((m->m_flags & M_PKTHDR) != 0, ("%s: !PKTHDR", __func__)); 1723 1724 /* While there's more data, get a new mbuf, tack it on, and fill it */ 1725 remain = m->m_pkthdr.len; 1726 moff = 0; 1727 p = ⊤ 1728 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */ 1729 struct mbuf *n; 1730 1731 /* Get the next new mbuf */ 1732 n = m_getl(remain, how, m->m_type, top == NULL ? M_PKTHDR : 0, 1733 &nsize); 1734 if (n == NULL) 1735 goto nospace; 1736 if (top == NULL) 1737 if (!m_dup_pkthdr(n, m, how)) 1738 goto nospace0; 1739 1740 /* Link it into the new chain */ 1741 *p = n; 1742 p = &n->m_next; 1743 1744 /* Copy data from original mbuf(s) into new mbuf */ 1745 n->m_len = 0; 1746 while (n->m_len < nsize && m != NULL) { 1747 int chunk = min(nsize - n->m_len, m->m_len - moff); 1748 1749 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk); 1750 moff += chunk; 1751 n->m_len += chunk; 1752 remain -= chunk; 1753 if (moff == m->m_len) { 1754 m = m->m_next; 1755 moff = 0; 1756 } 1757 } 1758 1759 /* Check correct total mbuf length */ 1760 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL), 1761 ("%s: bogus m_pkthdr.len", __func__)); 1762 } 1763 return (top); 1764 1765 nospace: 1766 m_freem(top); 1767 nospace0: 1768 ++mbstat[mycpu->gd_cpuid].m_mcfail; 1769 return (NULL); 1770 } 1771 1772 /* 1773 * Copy the non-packet mbuf data chain into a new set of mbufs, including 1774 * copying any mbuf clusters. This is typically used to realign a data 1775 * chain by nfs_realign(). 1776 * 1777 * The original chain is left intact. how should be M_WAITOK or M_NOWAIT 1778 * and NULL can be returned if M_NOWAIT is passed. 1779 * 1780 * Be careful to use cluster mbufs, a large mbuf chain converted to non 1781 * cluster mbufs can exhaust our supply of mbufs. 1782 */ 1783 struct mbuf * 1784 m_dup_data(struct mbuf *m, int how) 1785 { 1786 struct mbuf **p, *n, *top = NULL; 1787 int mlen, moff, chunk, gsize, nsize; 1788 1789 /* 1790 * Degenerate case 1791 */ 1792 if (m == NULL) 1793 return (NULL); 1794 1795 /* 1796 * Optimize the mbuf allocation but do not get too carried away. 1797 */ 1798 if (m->m_next || m->m_len > MLEN) 1799 if (m->m_flags & M_EXT && m->m_ext.ext_size == MCLBYTES) 1800 gsize = MCLBYTES; 1801 else 1802 gsize = MJUMPAGESIZE; 1803 else 1804 gsize = MLEN; 1805 1806 /* Chain control */ 1807 p = ⊤ 1808 n = NULL; 1809 nsize = 0; 1810 1811 /* 1812 * Scan the mbuf chain until nothing is left, the new mbuf chain 1813 * will be allocated on the fly as needed. 1814 */ 1815 while (m) { 1816 mlen = m->m_len; 1817 moff = 0; 1818 1819 while (mlen) { 1820 KKASSERT(m->m_type == MT_DATA); 1821 if (n == NULL) { 1822 n = m_getl(gsize, how, MT_DATA, 0, &nsize); 1823 n->m_len = 0; 1824 if (n == NULL) 1825 goto nospace; 1826 *p = n; 1827 p = &n->m_next; 1828 } 1829 chunk = imin(mlen, nsize); 1830 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk); 1831 mlen -= chunk; 1832 moff += chunk; 1833 n->m_len += chunk; 1834 nsize -= chunk; 1835 if (nsize == 0) 1836 n = NULL; 1837 } 1838 m = m->m_next; 1839 } 1840 *p = NULL; 1841 return(top); 1842 nospace: 1843 *p = NULL; 1844 m_freem(top); 1845 ++mbstat[mycpu->gd_cpuid].m_mcfail; 1846 return (NULL); 1847 } 1848 1849 /* 1850 * Concatenate mbuf chain n to m. 1851 * Both chains must be of the same type (e.g. MT_DATA). 1852 * Any m_pkthdr is not updated. 1853 */ 1854 void 1855 m_cat(struct mbuf *m, struct mbuf *n) 1856 { 1857 m = m_last(m); 1858 while (n) { 1859 if (m->m_flags & M_EXT || 1860 m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) { 1861 /* just join the two chains */ 1862 m->m_next = n; 1863 return; 1864 } 1865 /* splat the data from one into the other */ 1866 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 1867 (u_int)n->m_len); 1868 m->m_len += n->m_len; 1869 n = m_free(n); 1870 } 1871 } 1872 1873 void 1874 m_adj(struct mbuf *mp, int req_len) 1875 { 1876 int len = req_len; 1877 struct mbuf *m; 1878 int count; 1879 1880 if ((m = mp) == NULL) 1881 return; 1882 if (len >= 0) { 1883 /* 1884 * Trim from head. 1885 */ 1886 while (m != NULL && len > 0) { 1887 if (m->m_len <= len) { 1888 len -= m->m_len; 1889 m->m_len = 0; 1890 m = m->m_next; 1891 } else { 1892 m->m_len -= len; 1893 m->m_data += len; 1894 len = 0; 1895 } 1896 } 1897 m = mp; 1898 if (mp->m_flags & M_PKTHDR) 1899 m->m_pkthdr.len -= (req_len - len); 1900 } else { 1901 /* 1902 * Trim from tail. Scan the mbuf chain, 1903 * calculating its length and finding the last mbuf. 1904 * If the adjustment only affects this mbuf, then just 1905 * adjust and return. Otherwise, rescan and truncate 1906 * after the remaining size. 1907 */ 1908 len = -len; 1909 count = 0; 1910 for (;;) { 1911 count += m->m_len; 1912 if (m->m_next == NULL) 1913 break; 1914 m = m->m_next; 1915 } 1916 if (m->m_len >= len) { 1917 m->m_len -= len; 1918 if (mp->m_flags & M_PKTHDR) 1919 mp->m_pkthdr.len -= len; 1920 return; 1921 } 1922 count -= len; 1923 if (count < 0) 1924 count = 0; 1925 /* 1926 * Correct length for chain is "count". 1927 * Find the mbuf with last data, adjust its length, 1928 * and toss data from remaining mbufs on chain. 1929 */ 1930 m = mp; 1931 if (m->m_flags & M_PKTHDR) 1932 m->m_pkthdr.len = count; 1933 for (; m; m = m->m_next) { 1934 if (m->m_len >= count) { 1935 m->m_len = count; 1936 break; 1937 } 1938 count -= m->m_len; 1939 } 1940 while (m->m_next) 1941 (m = m->m_next) ->m_len = 0; 1942 } 1943 } 1944 1945 /* 1946 * Set the m_data pointer of a newly-allocated mbuf 1947 * to place an object of the specified size at the 1948 * end of the mbuf, longword aligned. 1949 */ 1950 void 1951 m_align(struct mbuf *m, int len) 1952 { 1953 int adjust; 1954 1955 if (m->m_flags & M_EXT) 1956 adjust = m->m_ext.ext_size - len; 1957 else if (m->m_flags & M_PKTHDR) 1958 adjust = MHLEN - len; 1959 else 1960 adjust = MLEN - len; 1961 m->m_data += adjust &~ (sizeof(long)-1); 1962 } 1963 1964 /* 1965 * Create a writable copy of the mbuf chain. While doing this 1966 * we compact the chain with a goal of producing a chain with 1967 * at most two mbufs. The second mbuf in this chain is likely 1968 * to be a cluster. The primary purpose of this work is to create 1969 * a writable packet for encryption, compression, etc. The 1970 * secondary goal is to linearize the data so the data can be 1971 * passed to crypto hardware in the most efficient manner possible. 1972 */ 1973 struct mbuf * 1974 m_unshare(struct mbuf *m0, int how) 1975 { 1976 struct mbuf *m, *mprev; 1977 struct mbuf *n, *mfirst, *mlast; 1978 int len, off; 1979 1980 mprev = NULL; 1981 for (m = m0; m != NULL; m = mprev->m_next) { 1982 /* 1983 * Regular mbufs are ignored unless there's a cluster 1984 * in front of it that we can use to coalesce. We do 1985 * the latter mainly so later clusters can be coalesced 1986 * also w/o having to handle them specially (i.e. convert 1987 * mbuf+cluster -> cluster). This optimization is heavily 1988 * influenced by the assumption that we're running over 1989 * Ethernet where MCLBYTES is large enough that the max 1990 * packet size will permit lots of coalescing into a 1991 * single cluster. This in turn permits efficient 1992 * crypto operations, especially when using hardware. 1993 */ 1994 if ((m->m_flags & M_EXT) == 0) { 1995 if (mprev && (mprev->m_flags & M_EXT) && 1996 m->m_len <= M_TRAILINGSPACE(mprev)) { 1997 /* XXX: this ignores mbuf types */ 1998 memcpy(mtod(mprev, caddr_t) + mprev->m_len, 1999 mtod(m, caddr_t), m->m_len); 2000 mprev->m_len += m->m_len; 2001 mprev->m_next = m->m_next; /* unlink from chain */ 2002 m_free(m); /* reclaim mbuf */ 2003 } else { 2004 mprev = m; 2005 } 2006 continue; 2007 } 2008 /* 2009 * Writable mbufs are left alone (for now). 2010 */ 2011 if (M_WRITABLE(m)) { 2012 mprev = m; 2013 continue; 2014 } 2015 2016 /* 2017 * Not writable, replace with a copy or coalesce with 2018 * the previous mbuf if possible (since we have to copy 2019 * it anyway, we try to reduce the number of mbufs and 2020 * clusters so that future work is easier). 2021 */ 2022 KASSERT(m->m_flags & M_EXT, ("m_flags 0x%x", m->m_flags)); 2023 /* NB: we only coalesce into a cluster or larger */ 2024 if (mprev != NULL && (mprev->m_flags & M_EXT) && 2025 m->m_len <= M_TRAILINGSPACE(mprev)) { 2026 /* XXX: this ignores mbuf types */ 2027 memcpy(mtod(mprev, caddr_t) + mprev->m_len, 2028 mtod(m, caddr_t), m->m_len); 2029 mprev->m_len += m->m_len; 2030 mprev->m_next = m->m_next; /* unlink from chain */ 2031 m_free(m); /* reclaim mbuf */ 2032 continue; 2033 } 2034 2035 /* 2036 * Allocate new space to hold the copy... 2037 */ 2038 /* XXX why can M_PKTHDR be set past the first mbuf? */ 2039 if (mprev == NULL && (m->m_flags & M_PKTHDR)) { 2040 /* 2041 * NB: if a packet header is present we must 2042 * allocate the mbuf separately from any cluster 2043 * because M_MOVE_PKTHDR will smash the data 2044 * pointer and drop the M_EXT marker. 2045 */ 2046 MGETHDR(n, how, m->m_type); 2047 if (n == NULL) { 2048 m_freem(m0); 2049 return (NULL); 2050 } 2051 M_MOVE_PKTHDR(n, m); 2052 MCLGET(n, how); 2053 if ((n->m_flags & M_EXT) == 0) { 2054 m_free(n); 2055 m_freem(m0); 2056 return (NULL); 2057 } 2058 } else { 2059 n = m_getcl(how, m->m_type, m->m_flags); 2060 if (n == NULL) { 2061 m_freem(m0); 2062 return (NULL); 2063 } 2064 } 2065 /* 2066 * ... and copy the data. We deal with jumbo mbufs 2067 * (i.e. m_len > MCLBYTES) by splitting them into 2068 * clusters. We could just malloc a buffer and make 2069 * it external but too many device drivers don't know 2070 * how to break up the non-contiguous memory when 2071 * doing DMA. 2072 */ 2073 len = m->m_len; 2074 off = 0; 2075 mfirst = n; 2076 mlast = NULL; 2077 for (;;) { 2078 int cc = min(len, MCLBYTES); 2079 memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + off, cc); 2080 n->m_len = cc; 2081 if (mlast != NULL) 2082 mlast->m_next = n; 2083 mlast = n; 2084 2085 len -= cc; 2086 if (len <= 0) 2087 break; 2088 off += cc; 2089 2090 n = m_getcl(how, m->m_type, m->m_flags); 2091 if (n == NULL) { 2092 m_freem(mfirst); 2093 m_freem(m0); 2094 return (NULL); 2095 } 2096 } 2097 n->m_next = m->m_next; 2098 if (mprev == NULL) 2099 m0 = mfirst; /* new head of chain */ 2100 else 2101 mprev->m_next = mfirst; /* replace old mbuf */ 2102 m_free(m); /* release old mbuf */ 2103 mprev = mfirst; 2104 } 2105 return (m0); 2106 } 2107 2108 /* 2109 * Rearrange an mbuf chain so that len bytes are contiguous 2110 * and in the data area of an mbuf (so that mtod will work for a structure 2111 * of size len). Returns the resulting mbuf chain on success, frees it and 2112 * returns null on failure. If there is room, it will add up to 2113 * max_protohdr-len extra bytes to the contiguous region in an attempt to 2114 * avoid being called next time. 2115 */ 2116 struct mbuf * 2117 m_pullup(struct mbuf *n, int len) 2118 { 2119 struct mbuf *m; 2120 int count; 2121 int space; 2122 2123 /* 2124 * If first mbuf has no cluster, and has room for len bytes 2125 * without shifting current data, pullup into it, 2126 * otherwise allocate a new mbuf to prepend to the chain. 2127 */ 2128 if (!(n->m_flags & M_EXT) && 2129 n->m_data + len < &n->m_dat[MLEN] && 2130 n->m_next) { 2131 if (n->m_len >= len) 2132 return (n); 2133 m = n; 2134 n = n->m_next; 2135 len -= m->m_len; 2136 } else { 2137 if (len > MHLEN) 2138 goto bad; 2139 if (n->m_flags & M_PKTHDR) 2140 m = m_gethdr(M_NOWAIT, n->m_type); 2141 else 2142 m = m_get(M_NOWAIT, n->m_type); 2143 if (m == NULL) 2144 goto bad; 2145 m->m_len = 0; 2146 if (n->m_flags & M_PKTHDR) 2147 M_MOVE_PKTHDR(m, n); 2148 } 2149 space = &m->m_dat[MLEN] - (m->m_data + m->m_len); 2150 do { 2151 count = min(min(max(len, max_protohdr), space), n->m_len); 2152 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 2153 (unsigned)count); 2154 len -= count; 2155 m->m_len += count; 2156 n->m_len -= count; 2157 space -= count; 2158 if (n->m_len) 2159 n->m_data += count; 2160 else 2161 n = m_free(n); 2162 } while (len > 0 && n); 2163 if (len > 0) { 2164 m_free(m); 2165 goto bad; 2166 } 2167 m->m_next = n; 2168 return (m); 2169 bad: 2170 m_freem(n); 2171 ++mbstat[mycpu->gd_cpuid].m_mcfail; 2172 return (NULL); 2173 } 2174 2175 /* 2176 * Partition an mbuf chain in two pieces, returning the tail -- 2177 * all but the first len0 bytes. In case of failure, it returns NULL and 2178 * attempts to restore the chain to its original state. 2179 * 2180 * Note that the resulting mbufs might be read-only, because the new 2181 * mbuf can end up sharing an mbuf cluster with the original mbuf if 2182 * the "breaking point" happens to lie within a cluster mbuf. Use the 2183 * M_WRITABLE() macro to check for this case. 2184 */ 2185 struct mbuf * 2186 m_split(struct mbuf *m0, int len0, int wait) 2187 { 2188 struct mbuf *m, *n; 2189 unsigned len = len0, remain; 2190 2191 for (m = m0; m && len > m->m_len; m = m->m_next) 2192 len -= m->m_len; 2193 if (m == NULL) 2194 return (NULL); 2195 remain = m->m_len - len; 2196 if (m0->m_flags & M_PKTHDR) { 2197 n = m_gethdr(wait, m0->m_type); 2198 if (n == NULL) 2199 return (NULL); 2200 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif; 2201 n->m_pkthdr.len = m0->m_pkthdr.len - len0; 2202 m0->m_pkthdr.len = len0; 2203 if (m->m_flags & M_EXT) 2204 goto extpacket; 2205 if (remain > MHLEN) { 2206 /* m can't be the lead packet */ 2207 MH_ALIGN(n, 0); 2208 n->m_next = m_split(m, len, wait); 2209 if (n->m_next == NULL) { 2210 m_free(n); 2211 return (NULL); 2212 } else { 2213 n->m_len = 0; 2214 return (n); 2215 } 2216 } else 2217 MH_ALIGN(n, remain); 2218 } else if (remain == 0) { 2219 n = m->m_next; 2220 m->m_next = NULL; 2221 return (n); 2222 } else { 2223 n = m_get(wait, m->m_type); 2224 if (n == NULL) 2225 return (NULL); 2226 M_ALIGN(n, remain); 2227 } 2228 extpacket: 2229 if (m->m_flags & M_EXT) { 2230 KKASSERT((n->m_flags & M_EXT) == 0); 2231 n->m_data = m->m_data + len; 2232 m->m_ext.ext_ref(m->m_ext.ext_arg); 2233 n->m_ext = m->m_ext; 2234 n->m_flags |= m->m_flags & (M_EXT | M_EXT_CLUSTER); 2235 } else { 2236 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain); 2237 } 2238 n->m_len = remain; 2239 m->m_len = len; 2240 n->m_next = m->m_next; 2241 m->m_next = NULL; 2242 return (n); 2243 } 2244 2245 /* 2246 * Routine to copy from device local memory into mbufs. 2247 * Note: "offset" is ill-defined and always called as 0, so ignore it. 2248 */ 2249 struct mbuf * 2250 m_devget(char *buf, int len, int offset, struct ifnet *ifp) 2251 { 2252 struct mbuf *m, *mfirst = NULL, **mtail; 2253 int nsize, flags; 2254 2255 mtail = &mfirst; 2256 flags = M_PKTHDR; 2257 2258 while (len > 0) { 2259 m = m_getl(len, M_NOWAIT, MT_DATA, flags, &nsize); 2260 if (m == NULL) { 2261 m_freem(mfirst); 2262 return (NULL); 2263 } 2264 m->m_len = min(len, nsize); 2265 2266 if (flags & M_PKTHDR) { 2267 if (len + max_linkhdr <= nsize) 2268 m->m_data += max_linkhdr; 2269 m->m_pkthdr.rcvif = ifp; 2270 m->m_pkthdr.len = len; 2271 flags = 0; 2272 } 2273 2274 bcopy(buf, m->m_data, (unsigned)m->m_len); 2275 buf += m->m_len; 2276 len -= m->m_len; 2277 *mtail = m; 2278 mtail = &m->m_next; 2279 } 2280 2281 return (mfirst); 2282 } 2283 2284 /* 2285 * Routine to pad mbuf to the specified length 'padto'. 2286 */ 2287 int 2288 m_devpad(struct mbuf *m, int padto) 2289 { 2290 struct mbuf *last = NULL; 2291 int padlen; 2292 2293 if (padto <= m->m_pkthdr.len) 2294 return 0; 2295 2296 padlen = padto - m->m_pkthdr.len; 2297 2298 /* if there's only the packet-header and we can pad there, use it. */ 2299 if (m->m_pkthdr.len == m->m_len && M_TRAILINGSPACE(m) >= padlen) { 2300 last = m; 2301 } else { 2302 /* 2303 * Walk packet chain to find last mbuf. We will either 2304 * pad there, or append a new mbuf and pad it 2305 */ 2306 for (last = m; last->m_next != NULL; last = last->m_next) 2307 ; /* EMPTY */ 2308 2309 /* `last' now points to last in chain. */ 2310 if (M_TRAILINGSPACE(last) < padlen) { 2311 struct mbuf *n; 2312 2313 /* Allocate new empty mbuf, pad it. Compact later. */ 2314 MGET(n, M_NOWAIT, MT_DATA); 2315 if (n == NULL) 2316 return ENOBUFS; 2317 n->m_len = 0; 2318 last->m_next = n; 2319 last = n; 2320 } 2321 } 2322 KKASSERT(M_TRAILINGSPACE(last) >= padlen); 2323 KKASSERT(M_WRITABLE(last)); 2324 2325 /* Now zero the pad area */ 2326 bzero(mtod(last, char *) + last->m_len, padlen); 2327 last->m_len += padlen; 2328 m->m_pkthdr.len += padlen; 2329 return 0; 2330 } 2331 2332 /* 2333 * Copy data from a buffer back into the indicated mbuf chain, 2334 * starting "off" bytes from the beginning, extending the mbuf 2335 * chain if necessary. 2336 */ 2337 void 2338 m_copyback(struct mbuf *m0, int off, int len, caddr_t cp) 2339 { 2340 int mlen; 2341 struct mbuf *m = m0, *n; 2342 int totlen = 0; 2343 2344 if (m0 == NULL) 2345 return; 2346 while (off > (mlen = m->m_len)) { 2347 off -= mlen; 2348 totlen += mlen; 2349 if (m->m_next == NULL) { 2350 n = m_getclr(M_NOWAIT, m->m_type); 2351 if (n == NULL) 2352 goto out; 2353 n->m_len = min(MLEN, len + off); 2354 m->m_next = n; 2355 } 2356 m = m->m_next; 2357 } 2358 while (len > 0) { 2359 mlen = min (m->m_len - off, len); 2360 bcopy(cp, off + mtod(m, caddr_t), (unsigned)mlen); 2361 cp += mlen; 2362 len -= mlen; 2363 mlen += off; 2364 off = 0; 2365 totlen += mlen; 2366 if (len == 0) 2367 break; 2368 if (m->m_next == NULL) { 2369 n = m_get(M_NOWAIT, m->m_type); 2370 if (n == NULL) 2371 break; 2372 n->m_len = min(MLEN, len); 2373 m->m_next = n; 2374 } 2375 m = m->m_next; 2376 } 2377 out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen)) 2378 m->m_pkthdr.len = totlen; 2379 } 2380 2381 /* 2382 * Append the specified data to the indicated mbuf chain, 2383 * Extend the mbuf chain if the new data does not fit in 2384 * existing space. 2385 * 2386 * Return 1 if able to complete the job; otherwise 0. 2387 */ 2388 int 2389 m_append(struct mbuf *m0, int len, c_caddr_t cp) 2390 { 2391 struct mbuf *m, *n; 2392 int remainder, space; 2393 2394 for (m = m0; m->m_next != NULL; m = m->m_next) 2395 ; 2396 remainder = len; 2397 space = M_TRAILINGSPACE(m); 2398 if (space > 0) { 2399 /* 2400 * Copy into available space. 2401 */ 2402 if (space > remainder) 2403 space = remainder; 2404 bcopy(cp, mtod(m, caddr_t) + m->m_len, space); 2405 m->m_len += space; 2406 cp += space, remainder -= space; 2407 } 2408 while (remainder > 0) { 2409 /* 2410 * Allocate a new mbuf; could check space 2411 * and allocate a cluster instead. 2412 */ 2413 n = m_get(M_NOWAIT, m->m_type); 2414 if (n == NULL) 2415 break; 2416 n->m_len = min(MLEN, remainder); 2417 bcopy(cp, mtod(n, caddr_t), n->m_len); 2418 cp += n->m_len, remainder -= n->m_len; 2419 m->m_next = n; 2420 m = n; 2421 } 2422 if (m0->m_flags & M_PKTHDR) 2423 m0->m_pkthdr.len += len - remainder; 2424 return (remainder == 0); 2425 } 2426 2427 /* 2428 * Apply function f to the data in an mbuf chain starting "off" bytes from 2429 * the beginning, continuing for "len" bytes. 2430 */ 2431 int 2432 m_apply(struct mbuf *m, int off, int len, 2433 int (*f)(void *, void *, u_int), void *arg) 2434 { 2435 u_int count; 2436 int rval; 2437 2438 KASSERT(off >= 0, ("m_apply, negative off %d", off)); 2439 KASSERT(len >= 0, ("m_apply, negative len %d", len)); 2440 while (off > 0) { 2441 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain")); 2442 if (off < m->m_len) 2443 break; 2444 off -= m->m_len; 2445 m = m->m_next; 2446 } 2447 while (len > 0) { 2448 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain")); 2449 count = min(m->m_len - off, len); 2450 rval = (*f)(arg, mtod(m, caddr_t) + off, count); 2451 if (rval) 2452 return (rval); 2453 len -= count; 2454 off = 0; 2455 m = m->m_next; 2456 } 2457 return (0); 2458 } 2459 2460 /* 2461 * Return a pointer to mbuf/offset of location in mbuf chain. 2462 */ 2463 struct mbuf * 2464 m_getptr(struct mbuf *m, int loc, int *off) 2465 { 2466 2467 while (loc >= 0) { 2468 /* Normal end of search. */ 2469 if (m->m_len > loc) { 2470 *off = loc; 2471 return (m); 2472 } else { 2473 loc -= m->m_len; 2474 if (m->m_next == NULL) { 2475 if (loc == 0) { 2476 /* Point at the end of valid data. */ 2477 *off = m->m_len; 2478 return (m); 2479 } 2480 return (NULL); 2481 } 2482 m = m->m_next; 2483 } 2484 } 2485 return (NULL); 2486 } 2487 2488 void 2489 m_print(const struct mbuf *m) 2490 { 2491 int len; 2492 const struct mbuf *m2; 2493 char *hexstr; 2494 2495 len = m->m_pkthdr.len; 2496 m2 = m; 2497 hexstr = kmalloc(HEX_NCPYLEN(len), M_TEMP, M_ZERO | M_WAITOK); 2498 while (len) { 2499 kprintf("%p %s\n", m2, hexncpy(m2->m_data, m2->m_len, hexstr, 2500 HEX_NCPYLEN(m2->m_len), "-")); 2501 len -= m2->m_len; 2502 m2 = m2->m_next; 2503 } 2504 kfree(hexstr, M_TEMP); 2505 return; 2506 } 2507 2508 /* 2509 * "Move" mbuf pkthdr from "from" to "to". 2510 * "from" must have M_PKTHDR set, and "to" must be empty. 2511 */ 2512 void 2513 m_move_pkthdr(struct mbuf *to, struct mbuf *from) 2514 { 2515 KASSERT((to->m_flags & M_PKTHDR), ("m_move_pkthdr: not packet header")); 2516 2517 to->m_flags |= from->m_flags & M_COPYFLAGS; 2518 to->m_pkthdr = from->m_pkthdr; /* especially tags */ 2519 SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */ 2520 } 2521 2522 /* 2523 * Duplicate "from"'s mbuf pkthdr in "to". 2524 * "from" must have M_PKTHDR set, and "to" must be empty. 2525 * In particular, this does a deep copy of the packet tags. 2526 */ 2527 int 2528 m_dup_pkthdr(struct mbuf *to, const struct mbuf *from, int how) 2529 { 2530 KASSERT((to->m_flags & M_PKTHDR), ("m_dup_pkthdr: not packet header")); 2531 2532 to->m_flags = (from->m_flags & M_COPYFLAGS) | 2533 (to->m_flags & ~M_COPYFLAGS); 2534 to->m_pkthdr = from->m_pkthdr; 2535 SLIST_INIT(&to->m_pkthdr.tags); 2536 return (m_tag_copy_chain(to, from, how)); 2537 } 2538 2539 /* 2540 * Defragment a mbuf chain, returning the shortest possible 2541 * chain of mbufs and clusters. If allocation fails and 2542 * this cannot be completed, NULL will be returned, but 2543 * the passed in chain will be unchanged. Upon success, 2544 * the original chain will be freed, and the new chain 2545 * will be returned. 2546 * 2547 * If a non-packet header is passed in, the original 2548 * mbuf (chain?) will be returned unharmed. 2549 * 2550 * m_defrag_nofree doesn't free the passed in mbuf. 2551 */ 2552 struct mbuf * 2553 m_defrag(struct mbuf *m0, int how) 2554 { 2555 struct mbuf *m_new; 2556 2557 if ((m_new = m_defrag_nofree(m0, how)) == NULL) 2558 return (NULL); 2559 if (m_new != m0) 2560 m_freem(m0); 2561 return (m_new); 2562 } 2563 2564 struct mbuf * 2565 m_defrag_nofree(struct mbuf *m0, int how) 2566 { 2567 struct mbuf *m_new = NULL, *m_final = NULL; 2568 int progress = 0, length, nsize; 2569 2570 if (!(m0->m_flags & M_PKTHDR)) 2571 return (m0); 2572 2573 #ifdef MBUF_STRESS_TEST 2574 if (m_defragrandomfailures) { 2575 int temp = karc4random() & 0xff; 2576 if (temp == 0xba) 2577 goto nospace; 2578 } 2579 #endif 2580 2581 m_final = m_getl(m0->m_pkthdr.len, how, MT_DATA, M_PKTHDR, &nsize); 2582 if (m_final == NULL) 2583 goto nospace; 2584 m_final->m_len = 0; /* in case m0->m_pkthdr.len is zero */ 2585 2586 if (m_dup_pkthdr(m_final, m0, how) == 0) 2587 goto nospace; 2588 2589 m_new = m_final; 2590 2591 while (progress < m0->m_pkthdr.len) { 2592 length = m0->m_pkthdr.len - progress; 2593 if (length > MCLBYTES) 2594 length = MCLBYTES; 2595 2596 if (m_new == NULL) { 2597 m_new = m_getl(length, how, MT_DATA, 0, &nsize); 2598 if (m_new == NULL) 2599 goto nospace; 2600 } 2601 2602 m_copydata(m0, progress, length, mtod(m_new, caddr_t)); 2603 progress += length; 2604 m_new->m_len = length; 2605 if (m_new != m_final) 2606 m_cat(m_final, m_new); 2607 m_new = NULL; 2608 } 2609 if (m0->m_next == NULL) 2610 m_defraguseless++; 2611 m_defragpackets++; 2612 m_defragbytes += m_final->m_pkthdr.len; 2613 return (m_final); 2614 nospace: 2615 m_defragfailure++; 2616 if (m_new) 2617 m_free(m_new); 2618 m_freem(m_final); 2619 return (NULL); 2620 } 2621 2622 /* 2623 * Move data from uio into mbufs. 2624 */ 2625 struct mbuf * 2626 m_uiomove(struct uio *uio) 2627 { 2628 struct mbuf *m; /* current working mbuf */ 2629 struct mbuf *head = NULL; /* result mbuf chain */ 2630 struct mbuf **mp = &head; 2631 int flags = M_PKTHDR; 2632 int nsize; 2633 int error; 2634 int resid; 2635 2636 do { 2637 if (uio->uio_resid > INT_MAX) 2638 resid = INT_MAX; 2639 else 2640 resid = (int)uio->uio_resid; 2641 m = m_getl(resid, M_WAITOK, MT_DATA, flags, &nsize); 2642 if (flags) { 2643 m->m_pkthdr.len = 0; 2644 /* Leave room for protocol headers. */ 2645 if (resid < MHLEN) 2646 MH_ALIGN(m, resid); 2647 flags = 0; 2648 } 2649 m->m_len = imin(nsize, resid); 2650 error = uiomove(mtod(m, caddr_t), m->m_len, uio); 2651 if (error) { 2652 m_free(m); 2653 goto failed; 2654 } 2655 *mp = m; 2656 mp = &m->m_next; 2657 head->m_pkthdr.len += m->m_len; 2658 } while (uio->uio_resid > 0); 2659 2660 return (head); 2661 2662 failed: 2663 m_freem(head); 2664 return (NULL); 2665 } 2666 2667 struct mbuf * 2668 m_last(struct mbuf *m) 2669 { 2670 while (m->m_next) 2671 m = m->m_next; 2672 return (m); 2673 } 2674 2675 /* 2676 * Return the number of bytes in an mbuf chain. 2677 * If lastm is not NULL, also return the last mbuf. 2678 */ 2679 u_int 2680 m_lengthm(struct mbuf *m, struct mbuf **lastm) 2681 { 2682 u_int len = 0; 2683 struct mbuf *prev = m; 2684 2685 while (m) { 2686 len += m->m_len; 2687 prev = m; 2688 m = m->m_next; 2689 } 2690 if (lastm != NULL) 2691 *lastm = prev; 2692 return (len); 2693 } 2694 2695 /* 2696 * Like m_lengthm(), except also keep track of mbuf usage. 2697 */ 2698 u_int 2699 m_countm(struct mbuf *m, struct mbuf **lastm, u_int *pmbcnt) 2700 { 2701 u_int len = 0, mbcnt = 0; 2702 struct mbuf *prev = m; 2703 2704 while (m) { 2705 len += m->m_len; 2706 mbcnt += MSIZE; 2707 if (m->m_flags & M_EXT) 2708 mbcnt += m->m_ext.ext_size; 2709 prev = m; 2710 m = m->m_next; 2711 } 2712 if (lastm != NULL) 2713 *lastm = prev; 2714 *pmbcnt = mbcnt; 2715 return (len); 2716 } 2717