kern_mbuf.c revision 253204
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 253204 2013-07-11 12:46:35Z andre $"); 30 31#include "opt_param.h" 32 33#include <sys/param.h> 34#include <sys/malloc.h> 35#include <sys/systm.h> 36#include <sys/mbuf.h> 37#include <sys/domain.h> 38#include <sys/eventhandler.h> 39#include <sys/kernel.h> 40#include <sys/protosw.h> 41#include <sys/smp.h> 42#include <sys/sysctl.h> 43 44#include <security/mac/mac_framework.h> 45 46#include <vm/vm.h> 47#include <vm/vm_extern.h> 48#include <vm/vm_kern.h> 49#include <vm/vm_page.h> 50#include <vm/vm_map.h> 51#include <vm/uma.h> 52#include <vm/uma_int.h> 53#include <vm/uma_dbg.h> 54 55/* 56 * In FreeBSD, Mbufs and Mbuf Clusters are allocated from UMA 57 * Zones. 58 * 59 * Mbuf Clusters (2K, contiguous) are allocated from the Cluster 60 * Zone. The Zone can be capped at kern.ipc.nmbclusters, if the 61 * administrator so desires. 62 * 63 * Mbufs are allocated from a UMA Master Zone called the Mbuf 64 * Zone. 65 * 66 * Additionally, FreeBSD provides a Packet Zone, which it 67 * configures as a Secondary Zone to the Mbuf Master Zone, 68 * thus sharing backend Slab kegs with the Mbuf Master Zone. 69 * 70 * Thus common-case allocations and locking are simplified: 71 * 72 * m_clget() m_getcl() 73 * | | 74 * | .------------>[(Packet Cache)] m_get(), m_gethdr() 75 * | | [ Packet ] | 76 * [(Cluster Cache)] [ Secondary ] [ (Mbuf Cache) ] 77 * [ Cluster Zone ] [ Zone ] [ Mbuf Master Zone ] 78 * | \________ | 79 * [ Cluster Keg ] \ / 80 * | [ Mbuf Keg ] 81 * [ Cluster Slabs ] | 82 * | [ Mbuf Slabs ] 83 * \____________(VM)_________________/ 84 * 85 * 86 * Whenever an object is allocated with uma_zalloc() out of 87 * one of the Zones its _ctor_ function is executed. The same 88 * for any deallocation through uma_zfree() the _dtor_ function 89 * is executed. 90 * 91 * Caches are per-CPU and are filled from the Master Zone. 92 * 93 * Whenever an object is allocated from the underlying global 94 * memory pool it gets pre-initialized with the _zinit_ functions. 95 * When the Keg's are overfull objects get decomissioned with 96 * _zfini_ functions and free'd back to the global memory pool. 97 * 98 */ 99 100int nmbufs; /* limits number of mbufs */ 101int nmbclusters; /* limits number of mbuf clusters */ 102int nmbjumbop; /* limits number of page size jumbo clusters */ 103int nmbjumbo9; /* limits number of 9k jumbo clusters */ 104int nmbjumbo16; /* limits number of 16k jumbo clusters */ 105struct mbstat mbstat; 106 107static quad_t maxmbufmem; /* overall real memory limit for all mbufs */ 108 109SYSCTL_QUAD(_kern_ipc, OID_AUTO, maxmbufmem, CTLFLAG_RDTUN, &maxmbufmem, 0, 110 "Maximum real memory allocateable to various mbuf types"); 111 112/* 113 * tunable_mbinit() has to be run before any mbuf allocations are done. 114 */ 115static void 116tunable_mbinit(void *dummy) 117{ 118 quad_t realmem; 119 120 /* 121 * The default limit for all mbuf related memory is 1/2 of all 122 * available kernel memory (physical or kmem). 123 * At most it can be 3/4 of available kernel memory. 124 */ 125 realmem = qmin((quad_t)physmem * PAGE_SIZE, 126 vm_map_max(kmem_map) - vm_map_min(kmem_map)); 127 maxmbufmem = realmem / 2; 128 TUNABLE_QUAD_FETCH("kern.ipc.maxmbufmem", &maxmbufmem); 129 if (maxmbufmem > realmem / 4 * 3) 130 maxmbufmem = realmem / 4 * 3; 131 132 TUNABLE_INT_FETCH("kern.ipc.nmbclusters", &nmbclusters); 133 if (nmbclusters == 0) 134 nmbclusters = maxmbufmem / MCLBYTES / 4; 135 136 TUNABLE_INT_FETCH("kern.ipc.nmbjumbop", &nmbjumbop); 137 if (nmbjumbop == 0) 138 nmbjumbop = maxmbufmem / MJUMPAGESIZE / 4; 139 140 TUNABLE_INT_FETCH("kern.ipc.nmbjumbo9", &nmbjumbo9); 141 if (nmbjumbo9 == 0) 142 nmbjumbo9 = maxmbufmem / MJUM9BYTES / 6; 143 144 TUNABLE_INT_FETCH("kern.ipc.nmbjumbo16", &nmbjumbo16); 145 if (nmbjumbo16 == 0) 146 nmbjumbo16 = maxmbufmem / MJUM16BYTES / 6; 147 148 /* 149 * We need at least as many mbufs as we have clusters of 150 * the various types added together. 151 */ 152 TUNABLE_INT_FETCH("kern.ipc.nmbufs", &nmbufs); 153 if (nmbufs < nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16) 154 nmbufs = lmax(maxmbufmem / MSIZE / 5, 155 nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16); 156} 157SYSINIT(tunable_mbinit, SI_SUB_KMEM, SI_ORDER_MIDDLE, tunable_mbinit, NULL); 158 159static int 160sysctl_nmbclusters(SYSCTL_HANDLER_ARGS) 161{ 162 int error, newnmbclusters; 163 164 newnmbclusters = nmbclusters; 165 error = sysctl_handle_int(oidp, &newnmbclusters, 0, req); 166 if (error == 0 && req->newptr) { 167 if (newnmbclusters > nmbclusters && 168 nmbufs >= nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16) { 169 nmbclusters = newnmbclusters; 170 uma_zone_set_max(zone_clust, nmbclusters); 171 nmbclusters = uma_zone_get_max(zone_clust); 172 EVENTHANDLER_INVOKE(nmbclusters_change); 173 } else 174 error = EINVAL; 175 } 176 return (error); 177} 178SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbclusters, CTLTYPE_INT|CTLFLAG_RW, 179&nmbclusters, 0, sysctl_nmbclusters, "IU", 180 "Maximum number of mbuf clusters allowed"); 181 182static int 183sysctl_nmbjumbop(SYSCTL_HANDLER_ARGS) 184{ 185 int error, newnmbjumbop; 186 187 newnmbjumbop = nmbjumbop; 188 error = sysctl_handle_int(oidp, &newnmbjumbop, 0, req); 189 if (error == 0 && req->newptr) { 190 if (newnmbjumbop > nmbjumbop && 191 nmbufs >= nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16) { 192 nmbjumbop = newnmbjumbop; 193 uma_zone_set_max(zone_jumbop, nmbjumbop); 194 nmbjumbop = uma_zone_get_max(zone_jumbop); 195 } else 196 error = EINVAL; 197 } 198 return (error); 199} 200SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbop, CTLTYPE_INT|CTLFLAG_RW, 201&nmbjumbop, 0, sysctl_nmbjumbop, "IU", 202 "Maximum number of mbuf page size jumbo clusters allowed"); 203 204static int 205sysctl_nmbjumbo9(SYSCTL_HANDLER_ARGS) 206{ 207 int error, newnmbjumbo9; 208 209 newnmbjumbo9 = nmbjumbo9; 210 error = sysctl_handle_int(oidp, &newnmbjumbo9, 0, req); 211 if (error == 0 && req->newptr) { 212 if (newnmbjumbo9 > nmbjumbo9 && 213 nmbufs >= nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16) { 214 nmbjumbo9 = newnmbjumbo9; 215 uma_zone_set_max(zone_jumbo9, nmbjumbo9); 216 nmbjumbo9 = uma_zone_get_max(zone_jumbo9); 217 } else 218 error = EINVAL; 219 } 220 return (error); 221} 222SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbo9, CTLTYPE_INT|CTLFLAG_RW, 223&nmbjumbo9, 0, sysctl_nmbjumbo9, "IU", 224 "Maximum number of mbuf 9k jumbo clusters allowed"); 225 226static int 227sysctl_nmbjumbo16(SYSCTL_HANDLER_ARGS) 228{ 229 int error, newnmbjumbo16; 230 231 newnmbjumbo16 = nmbjumbo16; 232 error = sysctl_handle_int(oidp, &newnmbjumbo16, 0, req); 233 if (error == 0 && req->newptr) { 234 if (newnmbjumbo16 > nmbjumbo16 && 235 nmbufs >= nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16) { 236 nmbjumbo16 = newnmbjumbo16; 237 uma_zone_set_max(zone_jumbo16, nmbjumbo16); 238 nmbjumbo16 = uma_zone_get_max(zone_jumbo16); 239 } else 240 error = EINVAL; 241 } 242 return (error); 243} 244SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbo16, CTLTYPE_INT|CTLFLAG_RW, 245&nmbjumbo16, 0, sysctl_nmbjumbo16, "IU", 246 "Maximum number of mbuf 16k jumbo clusters allowed"); 247 248static int 249sysctl_nmbufs(SYSCTL_HANDLER_ARGS) 250{ 251 int error, newnmbufs; 252 253 newnmbufs = nmbufs; 254 error = sysctl_handle_int(oidp, &newnmbufs, 0, req); 255 if (error == 0 && req->newptr) { 256 if (newnmbufs > nmbufs) { 257 nmbufs = newnmbufs; 258 uma_zone_set_max(zone_mbuf, nmbufs); 259 nmbufs = uma_zone_get_max(zone_mbuf); 260 EVENTHANDLER_INVOKE(nmbufs_change); 261 } else 262 error = EINVAL; 263 } 264 return (error); 265} 266SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbufs, CTLTYPE_INT|CTLFLAG_RW, 267&nmbufs, 0, sysctl_nmbufs, "IU", 268 "Maximum number of mbufs allowed"); 269 270SYSCTL_STRUCT(_kern_ipc, OID_AUTO, mbstat, CTLFLAG_RD, &mbstat, mbstat, 271 "Mbuf general information and statistics"); 272 273/* 274 * Zones from which we allocate. 275 */ 276uma_zone_t zone_mbuf; 277uma_zone_t zone_clust; 278uma_zone_t zone_pack; 279uma_zone_t zone_jumbop; 280uma_zone_t zone_jumbo9; 281uma_zone_t zone_jumbo16; 282uma_zone_t zone_ext_refcnt; 283 284/* 285 * Local prototypes. 286 */ 287static int mb_ctor_mbuf(void *, int, void *, int); 288static int mb_ctor_clust(void *, int, void *, int); 289static int mb_ctor_pack(void *, int, void *, int); 290static void mb_dtor_mbuf(void *, int, void *); 291static void mb_dtor_clust(void *, int, void *); 292static void mb_dtor_pack(void *, int, void *); 293static int mb_zinit_pack(void *, int, int); 294static void mb_zfini_pack(void *, int); 295 296static void mb_reclaim(void *); 297static void *mbuf_jumbo_alloc(uma_zone_t, int, uint8_t *, int); 298 299/* Ensure that MSIZE is a power of 2. */ 300CTASSERT((((MSIZE - 1) ^ MSIZE) + 1) >> 1 == MSIZE); 301 302/* 303 * Initialize FreeBSD Network buffer allocation. 304 */ 305static void 306mbuf_init(void *dummy) 307{ 308 309 /* 310 * Configure UMA zones for Mbufs, Clusters, and Packets. 311 */ 312 zone_mbuf = uma_zcreate(MBUF_MEM_NAME, MSIZE, 313 mb_ctor_mbuf, mb_dtor_mbuf, 314#ifdef INVARIANTS 315 trash_init, trash_fini, 316#else 317 NULL, NULL, 318#endif 319 MSIZE - 1, UMA_ZONE_MAXBUCKET); 320 if (nmbufs > 0) 321 nmbufs = uma_zone_set_max(zone_mbuf, nmbufs); 322 uma_zone_set_warning(zone_mbuf, "kern.ipc.nmbufs limit reached"); 323 324 zone_clust = uma_zcreate(MBUF_CLUSTER_MEM_NAME, MCLBYTES, 325 mb_ctor_clust, mb_dtor_clust, 326#ifdef INVARIANTS 327 trash_init, trash_fini, 328#else 329 NULL, NULL, 330#endif 331 UMA_ALIGN_PTR, UMA_ZONE_REFCNT); 332 if (nmbclusters > 0) 333 nmbclusters = uma_zone_set_max(zone_clust, nmbclusters); 334 uma_zone_set_warning(zone_clust, "kern.ipc.nmbclusters limit reached"); 335 336 zone_pack = uma_zsecond_create(MBUF_PACKET_MEM_NAME, mb_ctor_pack, 337 mb_dtor_pack, mb_zinit_pack, mb_zfini_pack, zone_mbuf); 338 339 /* Make jumbo frame zone too. Page size, 9k and 16k. */ 340 zone_jumbop = uma_zcreate(MBUF_JUMBOP_MEM_NAME, MJUMPAGESIZE, 341 mb_ctor_clust, mb_dtor_clust, 342#ifdef INVARIANTS 343 trash_init, trash_fini, 344#else 345 NULL, NULL, 346#endif 347 UMA_ALIGN_PTR, UMA_ZONE_REFCNT); 348 if (nmbjumbop > 0) 349 nmbjumbop = uma_zone_set_max(zone_jumbop, nmbjumbop); 350 uma_zone_set_warning(zone_jumbop, "kern.ipc.nmbjumbop limit reached"); 351 352 zone_jumbo9 = uma_zcreate(MBUF_JUMBO9_MEM_NAME, MJUM9BYTES, 353 mb_ctor_clust, mb_dtor_clust, 354#ifdef INVARIANTS 355 trash_init, trash_fini, 356#else 357 NULL, NULL, 358#endif 359 UMA_ALIGN_PTR, UMA_ZONE_REFCNT); 360 uma_zone_set_allocf(zone_jumbo9, mbuf_jumbo_alloc); 361 if (nmbjumbo9 > 0) 362 nmbjumbo9 = uma_zone_set_max(zone_jumbo9, nmbjumbo9); 363 uma_zone_set_warning(zone_jumbo9, "kern.ipc.nmbjumbo9 limit reached"); 364 365 zone_jumbo16 = uma_zcreate(MBUF_JUMBO16_MEM_NAME, MJUM16BYTES, 366 mb_ctor_clust, mb_dtor_clust, 367#ifdef INVARIANTS 368 trash_init, trash_fini, 369#else 370 NULL, NULL, 371#endif 372 UMA_ALIGN_PTR, UMA_ZONE_REFCNT); 373 uma_zone_set_allocf(zone_jumbo16, mbuf_jumbo_alloc); 374 if (nmbjumbo16 > 0) 375 nmbjumbo16 = uma_zone_set_max(zone_jumbo16, nmbjumbo16); 376 uma_zone_set_warning(zone_jumbo16, "kern.ipc.nmbjumbo16 limit reached"); 377 378 zone_ext_refcnt = uma_zcreate(MBUF_EXTREFCNT_MEM_NAME, sizeof(u_int), 379 NULL, NULL, 380 NULL, NULL, 381 UMA_ALIGN_PTR, UMA_ZONE_ZINIT); 382 383 /* uma_prealloc() goes here... */ 384 385 /* 386 * Hook event handler for low-memory situation, used to 387 * drain protocols and push data back to the caches (UMA 388 * later pushes it back to VM). 389 */ 390 EVENTHANDLER_REGISTER(vm_lowmem, mb_reclaim, NULL, 391 EVENTHANDLER_PRI_FIRST); 392 393 /* 394 * [Re]set counters and local statistics knobs. 395 * XXX Some of these should go and be replaced, but UMA stat 396 * gathering needs to be revised. 397 */ 398 mbstat.m_mbufs = 0; 399 mbstat.m_mclusts = 0; 400 mbstat.m_drain = 0; 401 mbstat.m_msize = MSIZE; 402 mbstat.m_mclbytes = MCLBYTES; 403 mbstat.m_minclsize = MINCLSIZE; 404 mbstat.m_mlen = MLEN; 405 mbstat.m_mhlen = MHLEN; 406 mbstat.m_numtypes = MT_NTYPES; 407 408 mbstat.m_mcfail = mbstat.m_mpfail = 0; 409 mbstat.sf_iocnt = 0; 410 mbstat.sf_allocwait = mbstat.sf_allocfail = 0; 411} 412SYSINIT(mbuf, SI_SUB_MBUF, SI_ORDER_FIRST, mbuf_init, NULL); 413 414/* 415 * UMA backend page allocator for the jumbo frame zones. 416 * 417 * Allocates kernel virtual memory that is backed by contiguous physical 418 * pages. 419 */ 420static void * 421mbuf_jumbo_alloc(uma_zone_t zone, int bytes, uint8_t *flags, int wait) 422{ 423 424 /* Inform UMA that this allocator uses kernel_map/object. */ 425 *flags = UMA_SLAB_KERNEL; 426 return ((void *)kmem_alloc_contig(kernel_map, bytes, wait, 427 (vm_paddr_t)0, ~(vm_paddr_t)0, 1, 0, VM_MEMATTR_DEFAULT)); 428} 429 430/* 431 * Constructor for Mbuf master zone. 432 * 433 * The 'arg' pointer points to a mb_args structure which 434 * contains call-specific information required to support the 435 * mbuf allocation API. See mbuf.h. 436 */ 437static int 438mb_ctor_mbuf(void *mem, int size, void *arg, int how) 439{ 440 struct mbuf *m; 441 struct mb_args *args; 442#ifdef MAC 443 int error; 444#endif 445 int flags; 446 short type; 447 448#ifdef INVARIANTS 449 trash_ctor(mem, size, arg, how); 450#endif 451 m = (struct mbuf *)mem; 452 args = (struct mb_args *)arg; 453 flags = args->flags; 454 type = args->type; 455 456 /* 457 * The mbuf is initialized later. The caller has the 458 * responsibility to set up any MAC labels too. 459 */ 460 if (type == MT_NOINIT) 461 return (0); 462 463 m->m_next = NULL; 464 m->m_nextpkt = NULL; 465 m->m_len = 0; 466 m->m_flags = flags; 467 m->m_type = type; 468 if (flags & M_PKTHDR) { 469 m->m_data = m->m_pktdat; 470 m->m_pkthdr.rcvif = NULL; 471 m->m_pkthdr.header = NULL; 472 m->m_pkthdr.len = 0; 473 m->m_pkthdr.csum_flags = 0; 474 m->m_pkthdr.csum_data = 0; 475 m->m_pkthdr.tso_segsz = 0; 476 m->m_pkthdr.ether_vtag = 0; 477 m->m_pkthdr.flowid = 0; 478 m->m_pkthdr.fibnum = 0; 479 SLIST_INIT(&m->m_pkthdr.tags); 480#ifdef MAC 481 /* If the label init fails, fail the alloc */ 482 error = mac_mbuf_init(m, how); 483 if (error) 484 return (error); 485#endif 486 } else 487 m->m_data = m->m_dat; 488 return (0); 489} 490 491/* 492 * The Mbuf master zone destructor. 493 */ 494static void 495mb_dtor_mbuf(void *mem, int size, void *arg) 496{ 497 struct mbuf *m; 498 unsigned long flags; 499 500 m = (struct mbuf *)mem; 501 flags = (unsigned long)arg; 502 503 if ((flags & MB_NOTAGS) == 0 && (m->m_flags & M_PKTHDR) != 0) 504 m_tag_delete_chain(m, NULL); 505 KASSERT((m->m_flags & M_EXT) == 0, ("%s: M_EXT set", __func__)); 506 KASSERT((m->m_flags & M_NOFREE) == 0, ("%s: M_NOFREE set", __func__)); 507#ifdef INVARIANTS 508 trash_dtor(mem, size, arg); 509#endif 510} 511 512/* 513 * The Mbuf Packet zone destructor. 514 */ 515static void 516mb_dtor_pack(void *mem, int size, void *arg) 517{ 518 struct mbuf *m; 519 520 m = (struct mbuf *)mem; 521 if ((m->m_flags & M_PKTHDR) != 0) 522 m_tag_delete_chain(m, NULL); 523 524 /* Make sure we've got a clean cluster back. */ 525 KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__)); 526 KASSERT(m->m_ext.ext_buf != NULL, ("%s: ext_buf == NULL", __func__)); 527 KASSERT(m->m_ext.ext_free == NULL, ("%s: ext_free != NULL", __func__)); 528 KASSERT(m->m_ext.ext_arg1 == NULL, ("%s: ext_arg1 != NULL", __func__)); 529 KASSERT(m->m_ext.ext_arg2 == NULL, ("%s: ext_arg2 != NULL", __func__)); 530 KASSERT(m->m_ext.ext_size == MCLBYTES, ("%s: ext_size != MCLBYTES", __func__)); 531 KASSERT(m->m_ext.ext_type == EXT_PACKET, ("%s: ext_type != EXT_PACKET", __func__)); 532 KASSERT(*m->m_ext.ref_cnt == 1, ("%s: ref_cnt != 1", __func__)); 533#ifdef INVARIANTS 534 trash_dtor(m->m_ext.ext_buf, MCLBYTES, arg); 535#endif 536 /* 537 * If there are processes blocked on zone_clust, waiting for pages 538 * to be freed up, * cause them to be woken up by draining the 539 * packet zone. We are exposed to a race here * (in the check for 540 * the UMA_ZFLAG_FULL) where we might miss the flag set, but that 541 * is deliberate. We don't want to acquire the zone lock for every 542 * mbuf free. 543 */ 544 if (uma_zone_exhausted_nolock(zone_clust)) 545 zone_drain(zone_pack); 546} 547 548/* 549 * The Cluster and Jumbo[PAGESIZE|9|16] zone constructor. 550 * 551 * Here the 'arg' pointer points to the Mbuf which we 552 * are configuring cluster storage for. If 'arg' is 553 * empty we allocate just the cluster without setting 554 * the mbuf to it. See mbuf.h. 555 */ 556static int 557mb_ctor_clust(void *mem, int size, void *arg, int how) 558{ 559 struct mbuf *m; 560 u_int *refcnt; 561 int type; 562 uma_zone_t zone; 563 564#ifdef INVARIANTS 565 trash_ctor(mem, size, arg, how); 566#endif 567 switch (size) { 568 case MCLBYTES: 569 type = EXT_CLUSTER; 570 zone = zone_clust; 571 break; 572#if MJUMPAGESIZE != MCLBYTES 573 case MJUMPAGESIZE: 574 type = EXT_JUMBOP; 575 zone = zone_jumbop; 576 break; 577#endif 578 case MJUM9BYTES: 579 type = EXT_JUMBO9; 580 zone = zone_jumbo9; 581 break; 582 case MJUM16BYTES: 583 type = EXT_JUMBO16; 584 zone = zone_jumbo16; 585 break; 586 default: 587 panic("unknown cluster size"); 588 break; 589 } 590 591 m = (struct mbuf *)arg; 592 refcnt = uma_find_refcnt(zone, mem); 593 *refcnt = 1; 594 if (m != NULL) { 595 m->m_ext.ext_buf = (caddr_t)mem; 596 m->m_data = m->m_ext.ext_buf; 597 m->m_flags |= M_EXT; 598 m->m_ext.ext_free = NULL; 599 m->m_ext.ext_arg1 = NULL; 600 m->m_ext.ext_arg2 = NULL; 601 m->m_ext.ext_size = size; 602 m->m_ext.ext_type = type; 603 m->m_ext.ref_cnt = refcnt; 604 } 605 606 return (0); 607} 608 609/* 610 * The Mbuf Cluster zone destructor. 611 */ 612static void 613mb_dtor_clust(void *mem, int size, void *arg) 614{ 615#ifdef INVARIANTS 616 uma_zone_t zone; 617 618 zone = m_getzone(size); 619 KASSERT(*(uma_find_refcnt(zone, mem)) <= 1, 620 ("%s: refcnt incorrect %u", __func__, 621 *(uma_find_refcnt(zone, mem))) ); 622 623 trash_dtor(mem, size, arg); 624#endif 625} 626 627/* 628 * The Packet secondary zone's init routine, executed on the 629 * object's transition from mbuf keg slab to zone cache. 630 */ 631static int 632mb_zinit_pack(void *mem, int size, int how) 633{ 634 struct mbuf *m; 635 636 m = (struct mbuf *)mem; /* m is virgin. */ 637 if (uma_zalloc_arg(zone_clust, m, how) == NULL || 638 m->m_ext.ext_buf == NULL) 639 return (ENOMEM); 640 m->m_ext.ext_type = EXT_PACKET; /* Override. */ 641#ifdef INVARIANTS 642 trash_init(m->m_ext.ext_buf, MCLBYTES, how); 643#endif 644 return (0); 645} 646 647/* 648 * The Packet secondary zone's fini routine, executed on the 649 * object's transition from zone cache to keg slab. 650 */ 651static void 652mb_zfini_pack(void *mem, int size) 653{ 654 struct mbuf *m; 655 656 m = (struct mbuf *)mem; 657#ifdef INVARIANTS 658 trash_fini(m->m_ext.ext_buf, MCLBYTES); 659#endif 660 uma_zfree_arg(zone_clust, m->m_ext.ext_buf, NULL); 661#ifdef INVARIANTS 662 trash_dtor(mem, size, NULL); 663#endif 664} 665 666/* 667 * The "packet" keg constructor. 668 */ 669static int 670mb_ctor_pack(void *mem, int size, void *arg, int how) 671{ 672 struct mbuf *m; 673 struct mb_args *args; 674#ifdef MAC 675 int error; 676#endif 677 int flags; 678 short type; 679 680 m = (struct mbuf *)mem; 681 args = (struct mb_args *)arg; 682 flags = args->flags; 683 type = args->type; 684 685#ifdef INVARIANTS 686 trash_ctor(m->m_ext.ext_buf, MCLBYTES, arg, how); 687#endif 688 m->m_next = NULL; 689 m->m_nextpkt = NULL; 690 m->m_data = m->m_ext.ext_buf; 691 m->m_len = 0; 692 m->m_flags = (flags | M_EXT); 693 m->m_type = type; 694 695 if (flags & M_PKTHDR) { 696 m->m_pkthdr.rcvif = NULL; 697 m->m_pkthdr.len = 0; 698 m->m_pkthdr.header = NULL; 699 m->m_pkthdr.csum_flags = 0; 700 m->m_pkthdr.csum_data = 0; 701 m->m_pkthdr.tso_segsz = 0; 702 m->m_pkthdr.ether_vtag = 0; 703 m->m_pkthdr.flowid = 0; 704 m->m_pkthdr.fibnum = 0; 705 SLIST_INIT(&m->m_pkthdr.tags); 706#ifdef MAC 707 /* If the label init fails, fail the alloc */ 708 error = mac_mbuf_init(m, how); 709 if (error) 710 return (error); 711#endif 712 } 713 /* m_ext is already initialized. */ 714 715 return (0); 716} 717 718int 719m_pkthdr_init(struct mbuf *m, int how) 720{ 721#ifdef MAC 722 int error; 723#endif 724 m->m_data = m->m_pktdat; 725 SLIST_INIT(&m->m_pkthdr.tags); 726 m->m_pkthdr.rcvif = NULL; 727 m->m_pkthdr.header = NULL; 728 m->m_pkthdr.len = 0; 729 m->m_pkthdr.flowid = 0; 730 m->m_pkthdr.fibnum = 0; 731 m->m_pkthdr.csum_flags = 0; 732 m->m_pkthdr.csum_data = 0; 733 m->m_pkthdr.tso_segsz = 0; 734 m->m_pkthdr.ether_vtag = 0; 735#ifdef MAC 736 /* If the label init fails, fail the alloc */ 737 error = mac_mbuf_init(m, how); 738 if (error) 739 return (error); 740#endif 741 742 return (0); 743} 744 745/* 746 * This is the protocol drain routine. 747 * 748 * No locks should be held when this is called. The drain routines have to 749 * presently acquire some locks which raises the possibility of lock order 750 * reversal. 751 */ 752static void 753mb_reclaim(void *junk) 754{ 755 struct domain *dp; 756 struct protosw *pr; 757 758 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK | WARN_PANIC, NULL, 759 "mb_reclaim()"); 760 761 for (dp = domains; dp != NULL; dp = dp->dom_next) 762 for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) 763 if (pr->pr_drain != NULL) 764 (*pr->pr_drain)(); 765} 766