kern_mbuf.c revision 162377
1/*- 2 * Copyright (c) 2004, 2005, 3 * Bosko Milekic <bmilekic@FreeBSD.org>. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice unmodified, this list of conditions and the following 10 * disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 25 * SUCH DAMAGE. 26 */ 27 28#include <sys/cdefs.h> 29__FBSDID("$FreeBSD: head/sys/kern/kern_mbuf.c 162377 2006-09-17 13:44:32Z andre $"); 30 31#include "opt_mac.h" 32#include "opt_param.h" 33 34#include <sys/param.h> 35#include <sys/mac.h> 36#include <sys/malloc.h> 37#include <sys/systm.h> 38#include <sys/mbuf.h> 39#include <sys/domain.h> 40#include <sys/eventhandler.h> 41#include <sys/kernel.h> 42#include <sys/protosw.h> 43#include <sys/smp.h> 44#include <sys/sysctl.h> 45 46#include <vm/vm.h> 47#include <vm/vm_page.h> 48#include <vm/uma.h> 49#include <vm/uma_int.h> 50#include <vm/uma_dbg.h> 51 52/* 53 * In FreeBSD, Mbufs and Mbuf Clusters are allocated from UMA 54 * Zones. 55 * 56 * Mbuf Clusters (2K, contiguous) are allocated from the Cluster 57 * Zone. The Zone can be capped at kern.ipc.nmbclusters, if the 58 * administrator so desires. 59 * 60 * Mbufs are allocated from a UMA Master Zone called the Mbuf 61 * Zone. 62 * 63 * Additionally, FreeBSD provides a Packet Zone, which it 64 * configures as a Secondary Zone to the Mbuf Master Zone, 65 * thus sharing backend Slab kegs with the Mbuf Master Zone. 66 * 67 * Thus common-case allocations and locking are simplified: 68 * 69 * m_clget() m_getcl() 70 * | | 71 * | .------------>[(Packet Cache)] m_get(), m_gethdr() 72 * | | [ Packet ] | 73 * [(Cluster Cache)] [ Secondary ] [ (Mbuf Cache) ] 74 * [ Cluster Zone ] [ Zone ] [ Mbuf Master Zone ] 75 * | \________ | 76 * [ Cluster Keg ] \ / 77 * | [ Mbuf Keg ] 78 * [ Cluster Slabs ] | 79 * | [ Mbuf Slabs ] 80 * \____________(VM)_________________/ 81 * 82 * 83 * Whenever an object is allocated with uma_zalloc() out of 84 * one of the Zones its _ctor_ function is executed. The same 85 * for any deallocation through uma_zfree() the _dtor_ function 86 * is executed. 87 * 88 * Caches are per-CPU and are filled from the Master Zone. 89 * 90 * Whenever an object is allocated from the underlying global 91 * memory pool it gets pre-initialized with the _zinit_ functions. 92 * When the Keg's are overfull objects get decomissioned with 93 * _zfini_ functions and free'd back to the global memory pool. 94 * 95 */ 96 97int nmbclusters; /* limits number of mbuf clusters */ 98int nmbjumbop; /* limits number of page size jumbo clusters */ 99int nmbjumbo9; /* limits number of 9k jumbo clusters */ 100int nmbjumbo16; /* limits number of 16k jumbo clusters */ 101struct mbstat mbstat; 102 103static void 104tunable_mbinit(void *dummy) 105{ 106 107 /* This has to be done before VM init. */ 108 nmbclusters = 1024 + maxusers * 64; 109 TUNABLE_INT_FETCH("kern.ipc.nmbclusters", &nmbclusters); 110} 111SYSINIT(tunable_mbinit, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_mbinit, NULL); 112 113/* XXX: These should be tuneables. Can't change UMA limits on the fly. */ 114static int 115sysctl_nmbclusters(SYSCTL_HANDLER_ARGS) 116{ 117 int error, newnmbclusters; 118 119 newnmbclusters = nmbclusters; 120 error = sysctl_handle_int(oidp, &newnmbclusters, sizeof(int), req); 121 if (error == 0 && req->newptr) { 122 if (newnmbclusters > nmbclusters) { 123 nmbclusters = newnmbclusters; 124 uma_zone_set_max(zone_clust, nmbclusters); 125 EVENTHANDLER_INVOKE(nmbclusters_change); 126 } else 127 error = EINVAL; 128 } 129 return (error); 130} 131SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbclusters, CTLTYPE_INT|CTLFLAG_RW, 132&nmbclusters, 0, sysctl_nmbclusters, "IU", 133 "Maximum number of mbuf clusters allowed"); 134SYSCTL_INT(_kern_ipc, OID_AUTO, nmbjumbop, CTLFLAG_RW, &nmbjumbop, 0, 135 "Maximum number of mbuf page size jumbo clusters allowed"); 136SYSCTL_INT(_kern_ipc, OID_AUTO, nmbjumbo9, CTLFLAG_RW, &nmbjumbo9, 0, 137 "Maximum number of mbuf 9k jumbo clusters allowed"); 138SYSCTL_INT(_kern_ipc, OID_AUTO, nmbjumbo16, CTLFLAG_RW, &nmbjumbo16, 0, 139 "Maximum number of mbuf 16k jumbo clusters allowed"); 140SYSCTL_STRUCT(_kern_ipc, OID_AUTO, mbstat, CTLFLAG_RD, &mbstat, mbstat, 141 "Mbuf general information and statistics"); 142 143/* 144 * Zones from which we allocate. 145 */ 146uma_zone_t zone_mbuf; 147uma_zone_t zone_clust; 148uma_zone_t zone_pack; 149uma_zone_t zone_jumbop; 150uma_zone_t zone_jumbo9; 151uma_zone_t zone_jumbo16; 152uma_zone_t zone_ext_refcnt; 153 154/* 155 * Local prototypes. 156 */ 157static int mb_ctor_mbuf(void *, int, void *, int); 158static int mb_ctor_clust(void *, int, void *, int); 159static int mb_ctor_pack(void *, int, void *, int); 160static void mb_dtor_mbuf(void *, int, void *); 161static void mb_dtor_clust(void *, int, void *); 162static void mb_dtor_pack(void *, int, void *); 163static int mb_zinit_pack(void *, int, int); 164static void mb_zfini_pack(void *, int); 165 166static void mb_reclaim(void *); 167static void mbuf_init(void *); 168 169/* Ensure that MSIZE doesn't break dtom() - it must be a power of 2 */ 170CTASSERT((((MSIZE - 1) ^ MSIZE) + 1) >> 1 == MSIZE); 171 172/* 173 * Initialize FreeBSD Network buffer allocation. 174 */ 175SYSINIT(mbuf, SI_SUB_MBUF, SI_ORDER_FIRST, mbuf_init, NULL) 176static void 177mbuf_init(void *dummy) 178{ 179 180 /* 181 * Configure UMA zones for Mbufs, Clusters, and Packets. 182 */ 183 zone_mbuf = uma_zcreate(MBUF_MEM_NAME, MSIZE, 184 mb_ctor_mbuf, mb_dtor_mbuf, 185#ifdef INVARIANTS 186 trash_init, trash_fini, 187#else 188 NULL, NULL, 189#endif 190 MSIZE - 1, UMA_ZONE_MAXBUCKET); 191 192 zone_clust = uma_zcreate(MBUF_CLUSTER_MEM_NAME, MCLBYTES, 193 mb_ctor_clust, mb_dtor_clust, 194#ifdef INVARIANTS 195 trash_init, trash_fini, 196#else 197 NULL, NULL, 198#endif 199 UMA_ALIGN_PTR, UMA_ZONE_REFCNT); 200 if (nmbclusters > 0) 201 uma_zone_set_max(zone_clust, nmbclusters); 202 203 zone_pack = uma_zsecond_create(MBUF_PACKET_MEM_NAME, mb_ctor_pack, 204 mb_dtor_pack, mb_zinit_pack, mb_zfini_pack, zone_mbuf); 205 206 /* Make jumbo frame zone too. Page size, 9k and 16k. */ 207 zone_jumbop = uma_zcreate(MBUF_JUMBOP_MEM_NAME, MJUMPAGESIZE, 208 mb_ctor_clust, mb_dtor_clust, 209#ifdef INVARIANTS 210 trash_init, trash_fini, 211#else 212 NULL, NULL, 213#endif 214 UMA_ALIGN_PTR, UMA_ZONE_REFCNT); 215 if (nmbjumbop > 0) 216 uma_zone_set_max(zone_jumbop, nmbjumbop); 217 218 zone_jumbo9 = uma_zcreate(MBUF_JUMBO9_MEM_NAME, MJUM9BYTES, 219 mb_ctor_clust, mb_dtor_clust, 220#ifdef INVARIANTS 221 trash_init, trash_fini, 222#else 223 NULL, NULL, 224#endif 225 UMA_ALIGN_PTR, UMA_ZONE_REFCNT); 226 if (nmbjumbo9 > 0) 227 uma_zone_set_max(zone_jumbo9, nmbjumbo9); 228 229 zone_jumbo16 = uma_zcreate(MBUF_JUMBO16_MEM_NAME, MJUM16BYTES, 230 mb_ctor_clust, mb_dtor_clust, 231#ifdef INVARIANTS 232 trash_init, trash_fini, 233#else 234 NULL, NULL, 235#endif 236 UMA_ALIGN_PTR, UMA_ZONE_REFCNT); 237 if (nmbjumbo16 > 0) 238 uma_zone_set_max(zone_jumbo16, nmbjumbo16); 239 240 zone_ext_refcnt = uma_zcreate(MBUF_EXTREFCNT_MEM_NAME, sizeof(u_int), 241 NULL, NULL, 242 NULL, NULL, 243 UMA_ALIGN_PTR, UMA_ZONE_ZINIT); 244 245 /* uma_prealloc() goes here... */ 246 247 /* 248 * Hook event handler for low-memory situation, used to 249 * drain protocols and push data back to the caches (UMA 250 * later pushes it back to VM). 251 */ 252 EVENTHANDLER_REGISTER(vm_lowmem, mb_reclaim, NULL, 253 EVENTHANDLER_PRI_FIRST); 254 255 /* 256 * [Re]set counters and local statistics knobs. 257 * XXX Some of these should go and be replaced, but UMA stat 258 * gathering needs to be revised. 259 */ 260 mbstat.m_mbufs = 0; 261 mbstat.m_mclusts = 0; 262 mbstat.m_drain = 0; 263 mbstat.m_msize = MSIZE; 264 mbstat.m_mclbytes = MCLBYTES; 265 mbstat.m_minclsize = MINCLSIZE; 266 mbstat.m_mlen = MLEN; 267 mbstat.m_mhlen = MHLEN; 268 mbstat.m_numtypes = MT_NTYPES; 269 270 mbstat.m_mcfail = mbstat.m_mpfail = 0; 271 mbstat.sf_iocnt = 0; 272 mbstat.sf_allocwait = mbstat.sf_allocfail = 0; 273} 274 275/* 276 * Constructor for Mbuf master zone. 277 * 278 * The 'arg' pointer points to a mb_args structure which 279 * contains call-specific information required to support the 280 * mbuf allocation API. See mbuf.h. 281 */ 282static int 283mb_ctor_mbuf(void *mem, int size, void *arg, int how) 284{ 285 struct mbuf *m; 286 struct mb_args *args; 287#ifdef MAC 288 int error; 289#endif 290 int flags; 291 short type; 292 293#ifdef INVARIANTS 294 trash_ctor(mem, size, arg, how); 295#endif 296 m = (struct mbuf *)mem; 297 args = (struct mb_args *)arg; 298 flags = args->flags; 299 type = args->type; 300 301 /* 302 * The mbuf is initialized later. The caller has the 303 * responsibility to set up any MAC labels too. 304 */ 305 if (type == MT_NOINIT) 306 return (0); 307 308 m->m_next = NULL; 309 m->m_nextpkt = NULL; 310 m->m_len = 0; 311 m->m_flags = flags; 312 m->m_type = type; 313 if (flags & M_PKTHDR) { 314 m->m_data = m->m_pktdat; 315 m->m_pkthdr.rcvif = NULL; 316 m->m_pkthdr.len = 0; 317 m->m_pkthdr.header = NULL; 318 m->m_pkthdr.csum_flags = 0; 319 m->m_pkthdr.csum_data = 0; 320 m->m_pkthdr.tso_segsz = 0; 321 m->m_pkthdr.ether_vtag = 0; 322 SLIST_INIT(&m->m_pkthdr.tags); 323#ifdef MAC 324 /* If the label init fails, fail the alloc */ 325 error = mac_init_mbuf(m, how); 326 if (error) 327 return (error); 328#endif 329 } else 330 m->m_data = m->m_dat; 331 return (0); 332} 333 334/* 335 * The Mbuf master zone destructor. 336 */ 337static void 338mb_dtor_mbuf(void *mem, int size, void *arg) 339{ 340 struct mbuf *m; 341 342 m = (struct mbuf *)mem; 343 if ((m->m_flags & M_PKTHDR) != 0) 344 m_tag_delete_chain(m, NULL); 345 KASSERT((m->m_flags & M_EXT) == 0, ("%s: M_EXT set", __func__)); 346#ifdef INVARIANTS 347 trash_dtor(mem, size, arg); 348#endif 349} 350 351/* 352 * The Mbuf Packet zone destructor. 353 */ 354static void 355mb_dtor_pack(void *mem, int size, void *arg) 356{ 357 struct mbuf *m; 358 359 m = (struct mbuf *)mem; 360 if ((m->m_flags & M_PKTHDR) != 0) 361 m_tag_delete_chain(m, NULL); 362 363 /* Make sure we've got a clean cluster back. */ 364 KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__)); 365 KASSERT(m->m_ext.ext_buf != NULL, ("%s: ext_buf == NULL", __func__)); 366 KASSERT(m->m_ext.ext_free == NULL, ("%s: ext_free != NULL", __func__)); 367 KASSERT(m->m_ext.ext_args == NULL, ("%s: ext_args != NULL", __func__)); 368 KASSERT(m->m_ext.ext_size == MCLBYTES, ("%s: ext_size != MCLBYTES", __func__)); 369 KASSERT(m->m_ext.ext_type == EXT_PACKET, ("%s: ext_type != EXT_PACKET", __func__)); 370 KASSERT(*m->m_ext.ref_cnt == 1, ("%s: ref_cnt != 1", __func__)); 371#ifdef INVARIANTS 372 trash_dtor(m->m_ext.ext_buf, MCLBYTES, arg); 373#endif 374} 375 376/* 377 * The Cluster and Jumbo[PAGESIZE|9|16] zone constructor. 378 * 379 * Here the 'arg' pointer points to the Mbuf which we 380 * are configuring cluster storage for. If 'arg' is 381 * empty we allocate just the cluster without setting 382 * the mbuf to it. See mbuf.h. 383 */ 384static int 385mb_ctor_clust(void *mem, int size, void *arg, int how) 386{ 387 struct mbuf *m; 388 u_int *refcnt; 389 int type = 0; 390 391#ifdef INVARIANTS 392 trash_ctor(mem, size, arg, how); 393#endif 394 m = (struct mbuf *)arg; 395 if (m != NULL) { 396 switch (size) { 397 case MCLBYTES: 398 type = EXT_CLUSTER; 399 break; 400#if MJUMPAGESIZE != MCLBYTES 401 case MJUMPAGESIZE: 402 type = EXT_JUMBOP; 403 break; 404#endif 405 case MJUM9BYTES: 406 type = EXT_JUMBO9; 407 break; 408 case MJUM16BYTES: 409 type = EXT_JUMBO16; 410 break; 411 default: 412 panic("unknown cluster size"); 413 break; 414 } 415 m->m_ext.ext_buf = (caddr_t)mem; 416 m->m_data = m->m_ext.ext_buf; 417 m->m_flags |= M_EXT; 418 m->m_ext.ext_free = NULL; 419 m->m_ext.ext_args = NULL; 420 m->m_ext.ext_size = size; 421 m->m_ext.ext_type = type; 422 m->m_ext.ref_cnt = uma_find_refcnt(zone_clust, mem); 423 *m->m_ext.ref_cnt = 1; 424 } else { 425 refcnt = uma_find_refcnt(zone_clust, mem); 426 *refcnt = 1; 427 } 428 return (0); 429} 430 431/* 432 * The Mbuf Cluster zone destructor. 433 */ 434static void 435mb_dtor_clust(void *mem, int size, void *arg) 436{ 437 438 KASSERT(*(uma_find_refcnt(zone_clust, mem)) <= 1, 439 ("%s: refcnt incorrect %u", __func__, 440 *(uma_find_refcnt(zone_clust, mem))) ); 441#ifdef INVARIANTS 442 trash_dtor(mem, size, arg); 443#endif 444} 445 446/* 447 * The Packet secondary zone's init routine, executed on the 448 * object's transition from mbuf keg slab to zone cache. 449 */ 450static int 451mb_zinit_pack(void *mem, int size, int how) 452{ 453 struct mbuf *m; 454 455 m = (struct mbuf *)mem; /* m is virgin. */ 456 if (uma_zalloc_arg(zone_clust, m, how) == NULL || 457 m->m_ext.ext_buf == NULL) 458 return (ENOMEM); 459 m->m_ext.ext_type = EXT_PACKET; /* Override. */ 460#ifdef INVARIANTS 461 trash_init(m->m_ext.ext_buf, MCLBYTES, how); 462#endif 463 return (0); 464} 465 466/* 467 * The Packet secondary zone's fini routine, executed on the 468 * object's transition from zone cache to keg slab. 469 */ 470static void 471mb_zfini_pack(void *mem, int size) 472{ 473 struct mbuf *m; 474 475 m = (struct mbuf *)mem; 476#ifdef INVARIANTS 477 trash_fini(m->m_ext.ext_buf, MCLBYTES); 478#endif 479 uma_zfree_arg(zone_clust, m->m_ext.ext_buf, NULL); 480#ifdef INVARIANTS 481 trash_dtor(mem, size, NULL); 482#endif 483} 484 485/* 486 * The "packet" keg constructor. 487 */ 488static int 489mb_ctor_pack(void *mem, int size, void *arg, int how) 490{ 491 struct mbuf *m; 492 struct mb_args *args; 493#ifdef MAC 494 int error; 495#endif 496 int flags; 497 short type; 498 499 m = (struct mbuf *)mem; 500 args = (struct mb_args *)arg; 501 flags = args->flags; 502 type = args->type; 503 504#ifdef INVARIANTS 505 trash_ctor(m->m_ext.ext_buf, MCLBYTES, arg, how); 506#endif 507 m->m_next = NULL; 508 m->m_nextpkt = NULL; 509 m->m_data = m->m_ext.ext_buf; 510 m->m_len = 0; 511 m->m_flags = (flags | M_EXT); 512 m->m_type = type; 513 514 if (flags & M_PKTHDR) { 515 m->m_pkthdr.rcvif = NULL; 516 m->m_pkthdr.len = 0; 517 m->m_pkthdr.header = NULL; 518 m->m_pkthdr.csum_flags = 0; 519 m->m_pkthdr.csum_data = 0; 520 m->m_pkthdr.tso_segsz = 0; 521 m->m_pkthdr.ether_vtag = 0; 522 SLIST_INIT(&m->m_pkthdr.tags); 523#ifdef MAC 524 /* If the label init fails, fail the alloc */ 525 error = mac_init_mbuf(m, how); 526 if (error) 527 return (error); 528#endif 529 } 530 /* m_ext is already initialized. */ 531 532 return (0); 533} 534 535/* 536 * This is the protocol drain routine. 537 * 538 * No locks should be held when this is called. The drain routines have to 539 * presently acquire some locks which raises the possibility of lock order 540 * reversal. 541 */ 542static void 543mb_reclaim(void *junk) 544{ 545 struct domain *dp; 546 struct protosw *pr; 547 548 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK | WARN_PANIC, NULL, 549 "mb_reclaim()"); 550 551 for (dp = domains; dp != NULL; dp = dp->dom_next) 552 for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) 553 if (pr->pr_drain != NULL) 554 (*pr->pr_drain)(); 555} 556