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("$FreeBSD$"); 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 && newnmbclusters != nmbclusters) { 167 if (newnmbclusters > nmbclusters && 168 nmbufs >= nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16) { 169 nmbclusters = newnmbclusters; 170 nmbclusters = uma_zone_set_max(zone_clust, nmbclusters); 171 EVENTHANDLER_INVOKE(nmbclusters_change); 172 } else 173 error = EINVAL; 174 } 175 return (error); 176} 177SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbclusters, CTLTYPE_INT|CTLFLAG_RW, 178&nmbclusters, 0, sysctl_nmbclusters, "IU", 179 "Maximum number of mbuf clusters allowed"); 180 181static int 182sysctl_nmbjumbop(SYSCTL_HANDLER_ARGS) 183{ 184 int error, newnmbjumbop; 185 186 newnmbjumbop = nmbjumbop; 187 error = sysctl_handle_int(oidp, &newnmbjumbop, 0, req); 188 if (error == 0 && req->newptr && newnmbjumbop != nmbjumbop) { 189 if (newnmbjumbop > nmbjumbop && 190 nmbufs >= nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16) { 191 nmbjumbop = newnmbjumbop; 192 nmbjumbop = uma_zone_set_max(zone_jumbop, nmbjumbop); 193 } else 194 error = EINVAL; 195 } 196 return (error); 197} 198SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbop, CTLTYPE_INT|CTLFLAG_RW, 199&nmbjumbop, 0, sysctl_nmbjumbop, "IU", 200 "Maximum number of mbuf page size jumbo clusters allowed"); 201 202static int 203sysctl_nmbjumbo9(SYSCTL_HANDLER_ARGS) 204{ 205 int error, newnmbjumbo9; 206 207 newnmbjumbo9 = nmbjumbo9; 208 error = sysctl_handle_int(oidp, &newnmbjumbo9, 0, req); 209 if (error == 0 && req->newptr && newnmbjumbo9 != nmbjumbo9) { 210 if (newnmbjumbo9 > nmbjumbo9 && 211 nmbufs >= nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16) { 212 nmbjumbo9 = newnmbjumbo9; 213 nmbjumbo9 = uma_zone_set_max(zone_jumbo9, nmbjumbo9); 214 } else 215 error = EINVAL; 216 } 217 return (error); 218} 219SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbo9, CTLTYPE_INT|CTLFLAG_RW, 220&nmbjumbo9, 0, sysctl_nmbjumbo9, "IU", 221 "Maximum number of mbuf 9k jumbo clusters allowed"); 222 223static int 224sysctl_nmbjumbo16(SYSCTL_HANDLER_ARGS) 225{ 226 int error, newnmbjumbo16; 227 228 newnmbjumbo16 = nmbjumbo16; 229 error = sysctl_handle_int(oidp, &newnmbjumbo16, 0, req); 230 if (error == 0 && req->newptr && newnmbjumbo16 != nmbjumbo16) { 231 if (newnmbjumbo16 > nmbjumbo16 && 232 nmbufs >= nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16) { 233 nmbjumbo16 = newnmbjumbo16; 234 nmbjumbo16 = uma_zone_set_max(zone_jumbo16, nmbjumbo16); 235 } else 236 error = EINVAL; 237 } 238 return (error); 239} 240SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbo16, CTLTYPE_INT|CTLFLAG_RW, 241&nmbjumbo16, 0, sysctl_nmbjumbo16, "IU", 242 "Maximum number of mbuf 16k jumbo clusters allowed"); 243 244static int 245sysctl_nmbufs(SYSCTL_HANDLER_ARGS) 246{ 247 int error, newnmbufs; 248 249 newnmbufs = nmbufs; 250 error = sysctl_handle_int(oidp, &newnmbufs, 0, req); 251 if (error == 0 && req->newptr && newnmbufs != nmbufs) { 252 if (newnmbufs > nmbufs) { 253 nmbufs = newnmbufs; 254 nmbufs = uma_zone_set_max(zone_mbuf, nmbufs); 255 EVENTHANDLER_INVOKE(nmbufs_change); 256 } else 257 error = EINVAL; 258 } 259 return (error); 260} 261SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbufs, CTLTYPE_INT|CTLFLAG_RW, 262&nmbufs, 0, sysctl_nmbufs, "IU", 263 "Maximum number of mbufs allowed"); 264 265SYSCTL_STRUCT(_kern_ipc, OID_AUTO, mbstat, CTLFLAG_RD, &mbstat, mbstat, 266 "Mbuf general information and statistics"); 267 268/* 269 * Zones from which we allocate. 270 */ 271uma_zone_t zone_mbuf; 272uma_zone_t zone_clust; 273uma_zone_t zone_pack; 274uma_zone_t zone_jumbop; 275uma_zone_t zone_jumbo9; 276uma_zone_t zone_jumbo16; 277uma_zone_t zone_ext_refcnt; 278 279/* 280 * Local prototypes. 281 */ 282static int mb_ctor_mbuf(void *, int, void *, int); 283static int mb_ctor_clust(void *, int, void *, int); 284static int mb_ctor_pack(void *, int, void *, int); 285static void mb_dtor_mbuf(void *, int, void *); 286static void mb_dtor_clust(void *, int, void *); 287static void mb_dtor_pack(void *, int, void *); 288static int mb_zinit_pack(void *, int, int); 289static void mb_zfini_pack(void *, int); 290 291static void mb_reclaim(void *); 292static void *mbuf_jumbo_alloc(uma_zone_t, int, uint8_t *, int); 293 294/* Ensure that MSIZE doesn't break dtom() - it must be a power of 2 */ 295CTASSERT((((MSIZE - 1) ^ MSIZE) + 1) >> 1 == MSIZE); 296 297/* 298 * Initialize FreeBSD Network buffer allocation. 299 */ 300static void 301mbuf_init(void *dummy) 302{ 303 304 /* 305 * Configure UMA zones for Mbufs, Clusters, and Packets. 306 */ 307 zone_mbuf = uma_zcreate(MBUF_MEM_NAME, MSIZE, 308 mb_ctor_mbuf, mb_dtor_mbuf, 309#ifdef INVARIANTS 310 trash_init, trash_fini, 311#else 312 NULL, NULL, 313#endif 314 MSIZE - 1, UMA_ZONE_MAXBUCKET); 315 if (nmbufs > 0) 316 nmbufs = uma_zone_set_max(zone_mbuf, nmbufs); 317 318 zone_clust = uma_zcreate(MBUF_CLUSTER_MEM_NAME, MCLBYTES, 319 mb_ctor_clust, mb_dtor_clust, 320#ifdef INVARIANTS 321 trash_init, trash_fini, 322#else 323 NULL, NULL, 324#endif 325 UMA_ALIGN_PTR, UMA_ZONE_REFCNT); 326 if (nmbclusters > 0) 327 nmbclusters = uma_zone_set_max(zone_clust, nmbclusters); 328 329 zone_pack = uma_zsecond_create(MBUF_PACKET_MEM_NAME, mb_ctor_pack, 330 mb_dtor_pack, mb_zinit_pack, mb_zfini_pack, zone_mbuf); 331 332 /* Make jumbo frame zone too. Page size, 9k and 16k. */ 333 zone_jumbop = uma_zcreate(MBUF_JUMBOP_MEM_NAME, MJUMPAGESIZE, 334 mb_ctor_clust, mb_dtor_clust, 335#ifdef INVARIANTS 336 trash_init, trash_fini, 337#else 338 NULL, NULL, 339#endif 340 UMA_ALIGN_PTR, UMA_ZONE_REFCNT); 341 if (nmbjumbop > 0) 342 nmbjumbop = uma_zone_set_max(zone_jumbop, nmbjumbop); 343 344 zone_jumbo9 = uma_zcreate(MBUF_JUMBO9_MEM_NAME, MJUM9BYTES, 345 mb_ctor_clust, mb_dtor_clust, 346#ifdef INVARIANTS 347 trash_init, trash_fini, 348#else 349 NULL, NULL, 350#endif 351 UMA_ALIGN_PTR, UMA_ZONE_REFCNT); 352 uma_zone_set_allocf(zone_jumbo9, mbuf_jumbo_alloc); 353 if (nmbjumbo9 > 0) 354 nmbjumbo9 = uma_zone_set_max(zone_jumbo9, nmbjumbo9); 355 356 zone_jumbo16 = uma_zcreate(MBUF_JUMBO16_MEM_NAME, MJUM16BYTES, 357 mb_ctor_clust, mb_dtor_clust, 358#ifdef INVARIANTS 359 trash_init, trash_fini, 360#else 361 NULL, NULL, 362#endif 363 UMA_ALIGN_PTR, UMA_ZONE_REFCNT); 364 uma_zone_set_allocf(zone_jumbo16, mbuf_jumbo_alloc); 365 if (nmbjumbo16 > 0) 366 nmbjumbo16 = uma_zone_set_max(zone_jumbo16, nmbjumbo16); 367 368 zone_ext_refcnt = uma_zcreate(MBUF_EXTREFCNT_MEM_NAME, sizeof(u_int), 369 NULL, NULL, 370 NULL, NULL, 371 UMA_ALIGN_PTR, UMA_ZONE_ZINIT); 372 373 /* uma_prealloc() goes here... */ 374 375 /* 376 * Hook event handler for low-memory situation, used to 377 * drain protocols and push data back to the caches (UMA 378 * later pushes it back to VM). 379 */ 380 EVENTHANDLER_REGISTER(vm_lowmem, mb_reclaim, NULL, 381 EVENTHANDLER_PRI_FIRST); 382 383 /* 384 * [Re]set counters and local statistics knobs. 385 * XXX Some of these should go and be replaced, but UMA stat 386 * gathering needs to be revised. 387 */ 388 mbstat.m_mbufs = 0; 389 mbstat.m_mclusts = 0; 390 mbstat.m_drain = 0; 391 mbstat.m_msize = MSIZE; 392 mbstat.m_mclbytes = MCLBYTES; 393 mbstat.m_minclsize = MINCLSIZE; 394 mbstat.m_mlen = MLEN; 395 mbstat.m_mhlen = MHLEN; 396 mbstat.m_numtypes = MT_NTYPES; 397 398 mbstat.m_mcfail = mbstat.m_mpfail = 0; 399 mbstat.sf_iocnt = 0; 400 mbstat.sf_allocwait = mbstat.sf_allocfail = 0; 401} 402SYSINIT(mbuf, SI_SUB_MBUF, SI_ORDER_FIRST, mbuf_init, NULL); 403 404/* 405 * UMA backend page allocator for the jumbo frame zones. 406 * 407 * Allocates kernel virtual memory that is backed by contiguous physical 408 * pages. 409 */ 410static void * 411mbuf_jumbo_alloc(uma_zone_t zone, int bytes, uint8_t *flags, int wait) 412{ 413 414 /* Inform UMA that this allocator uses kernel_map/object. */ 415 *flags = UMA_SLAB_KERNEL; 416 return ((void *)kmem_alloc_contig(kernel_map, bytes, wait, 417 (vm_paddr_t)0, ~(vm_paddr_t)0, 1, 0, VM_MEMATTR_DEFAULT)); 418} 419 420/* 421 * Constructor for Mbuf master zone. 422 * 423 * The 'arg' pointer points to a mb_args structure which 424 * contains call-specific information required to support the 425 * mbuf allocation API. See mbuf.h. 426 */ 427static int 428mb_ctor_mbuf(void *mem, int size, void *arg, int how) 429{ 430 struct mbuf *m; 431 struct mb_args *args; 432#ifdef MAC 433 int error; 434#endif 435 int flags; 436 short type; 437 438#ifdef INVARIANTS 439 trash_ctor(mem, size, arg, how); 440#endif 441 m = (struct mbuf *)mem; 442 args = (struct mb_args *)arg; 443 flags = args->flags; 444 type = args->type; 445 446 /* 447 * The mbuf is initialized later. The caller has the 448 * responsibility to set up any MAC labels too. 449 */ 450 if (type == MT_NOINIT) 451 return (0); 452 453 m->m_next = NULL; 454 m->m_nextpkt = NULL; 455 m->m_len = 0; 456 m->m_flags = flags; 457 m->m_type = type; 458 if (flags & M_PKTHDR) { 459 m->m_data = m->m_pktdat; 460 m->m_pkthdr.rcvif = NULL; 461 m->m_pkthdr.header = NULL; 462 m->m_pkthdr.len = 0; 463 m->m_pkthdr.csum_flags = 0; 464 m->m_pkthdr.csum_data = 0; 465 m->m_pkthdr.tso_segsz = 0; 466 m->m_pkthdr.ether_vtag = 0; 467 m->m_pkthdr.flowid = 0; 468 SLIST_INIT(&m->m_pkthdr.tags); 469#ifdef MAC 470 /* If the label init fails, fail the alloc */ 471 error = mac_mbuf_init(m, how); 472 if (error) 473 return (error); 474#endif 475 } else 476 m->m_data = m->m_dat; 477 return (0); 478} 479 480/* 481 * The Mbuf master zone destructor. 482 */ 483static void 484mb_dtor_mbuf(void *mem, int size, void *arg) 485{ 486 struct mbuf *m; 487 unsigned long flags; 488 489 m = (struct mbuf *)mem; 490 flags = (unsigned long)arg; 491 492 if ((flags & MB_NOTAGS) == 0 && (m->m_flags & M_PKTHDR) != 0) 493 m_tag_delete_chain(m, NULL); 494 KASSERT((m->m_flags & M_EXT) == 0, ("%s: M_EXT set", __func__)); 495 KASSERT((m->m_flags & M_NOFREE) == 0, ("%s: M_NOFREE set", __func__)); 496#ifdef INVARIANTS 497 trash_dtor(mem, size, arg); 498#endif 499} 500 501/* 502 * The Mbuf Packet zone destructor. 503 */ 504static void 505mb_dtor_pack(void *mem, int size, void *arg) 506{ 507 struct mbuf *m; 508 509 m = (struct mbuf *)mem; 510 if ((m->m_flags & M_PKTHDR) != 0) 511 m_tag_delete_chain(m, NULL); 512 513 /* Make sure we've got a clean cluster back. */ 514 KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__)); 515 KASSERT(m->m_ext.ext_buf != NULL, ("%s: ext_buf == NULL", __func__)); 516 KASSERT(m->m_ext.ext_free == NULL, ("%s: ext_free != NULL", __func__)); 517 KASSERT(m->m_ext.ext_arg1 == NULL, ("%s: ext_arg1 != NULL", __func__)); 518 KASSERT(m->m_ext.ext_arg2 == NULL, ("%s: ext_arg2 != NULL", __func__)); 519 KASSERT(m->m_ext.ext_size == MCLBYTES, ("%s: ext_size != MCLBYTES", __func__)); 520 KASSERT(m->m_ext.ext_type == EXT_PACKET, ("%s: ext_type != EXT_PACKET", __func__)); 521 KASSERT(*m->m_ext.ref_cnt == 1, ("%s: ref_cnt != 1", __func__)); 522#ifdef INVARIANTS 523 trash_dtor(m->m_ext.ext_buf, MCLBYTES, arg); 524#endif 525 /* 526 * If there are processes blocked on zone_clust, waiting for pages 527 * to be freed up, * cause them to be woken up by draining the 528 * packet zone. We are exposed to a race here * (in the check for 529 * the UMA_ZFLAG_FULL) where we might miss the flag set, but that 530 * is deliberate. We don't want to acquire the zone lock for every 531 * mbuf free. 532 */ 533 if (uma_zone_exhausted_nolock(zone_clust)) 534 zone_drain(zone_pack); 535} 536 537/* 538 * The Cluster and Jumbo[PAGESIZE|9|16] zone constructor. 539 * 540 * Here the 'arg' pointer points to the Mbuf which we 541 * are configuring cluster storage for. If 'arg' is 542 * empty we allocate just the cluster without setting 543 * the mbuf to it. See mbuf.h. 544 */ 545static int 546mb_ctor_clust(void *mem, int size, void *arg, int how) 547{ 548 struct mbuf *m; 549 u_int *refcnt; 550 int type; 551 uma_zone_t zone; 552 553#ifdef INVARIANTS 554 trash_ctor(mem, size, arg, how); 555#endif 556 switch (size) { 557 case MCLBYTES: 558 type = EXT_CLUSTER; 559 zone = zone_clust; 560 break; 561#if MJUMPAGESIZE != MCLBYTES 562 case MJUMPAGESIZE: 563 type = EXT_JUMBOP; 564 zone = zone_jumbop; 565 break; 566#endif 567 case MJUM9BYTES: 568 type = EXT_JUMBO9; 569 zone = zone_jumbo9; 570 break; 571 case MJUM16BYTES: 572 type = EXT_JUMBO16; 573 zone = zone_jumbo16; 574 break; 575 default: 576 panic("unknown cluster size"); 577 break; 578 } 579 580 m = (struct mbuf *)arg; 581 refcnt = uma_find_refcnt(zone, mem); 582 *refcnt = 1; 583 if (m != NULL) { 584 m->m_ext.ext_buf = (caddr_t)mem; 585 m->m_data = m->m_ext.ext_buf; 586 m->m_flags |= M_EXT; 587 m->m_ext.ext_free = NULL; 588 m->m_ext.ext_arg1 = NULL; 589 m->m_ext.ext_arg2 = NULL; 590 m->m_ext.ext_size = size; 591 m->m_ext.ext_type = type; 592 m->m_ext.ref_cnt = refcnt; 593 } 594 595 return (0); 596} 597 598/* 599 * The Mbuf Cluster zone destructor. 600 */ 601static void 602mb_dtor_clust(void *mem, int size, void *arg) 603{ 604#ifdef INVARIANTS 605 uma_zone_t zone; 606 607 zone = m_getzone(size); 608 KASSERT(*(uma_find_refcnt(zone, mem)) <= 1, 609 ("%s: refcnt incorrect %u", __func__, 610 *(uma_find_refcnt(zone, mem))) ); 611 612 trash_dtor(mem, size, arg); 613#endif 614} 615 616/* 617 * The Packet secondary zone's init routine, executed on the 618 * object's transition from mbuf keg slab to zone cache. 619 */ 620static int 621mb_zinit_pack(void *mem, int size, int how) 622{ 623 struct mbuf *m; 624 625 m = (struct mbuf *)mem; /* m is virgin. */ 626 if (uma_zalloc_arg(zone_clust, m, how) == NULL || 627 m->m_ext.ext_buf == NULL) 628 return (ENOMEM); 629 m->m_ext.ext_type = EXT_PACKET; /* Override. */ 630#ifdef INVARIANTS 631 trash_init(m->m_ext.ext_buf, MCLBYTES, how); 632#endif 633 return (0); 634} 635 636/* 637 * The Packet secondary zone's fini routine, executed on the 638 * object's transition from zone cache to keg slab. 639 */ 640static void 641mb_zfini_pack(void *mem, int size) 642{ 643 struct mbuf *m; 644 645 m = (struct mbuf *)mem; 646#ifdef INVARIANTS 647 trash_fini(m->m_ext.ext_buf, MCLBYTES); 648#endif 649 uma_zfree_arg(zone_clust, m->m_ext.ext_buf, NULL); 650#ifdef INVARIANTS 651 trash_dtor(mem, size, NULL); 652#endif 653} 654 655/* 656 * The "packet" keg constructor. 657 */ 658static int 659mb_ctor_pack(void *mem, int size, void *arg, int how) 660{ 661 struct mbuf *m; 662 struct mb_args *args; 663#ifdef MAC 664 int error; 665#endif 666 int flags; 667 short type; 668 669 m = (struct mbuf *)mem; 670 args = (struct mb_args *)arg; 671 flags = args->flags; 672 type = args->type; 673 674#ifdef INVARIANTS 675 trash_ctor(m->m_ext.ext_buf, MCLBYTES, arg, how); 676#endif 677 m->m_next = NULL; 678 m->m_nextpkt = NULL; 679 m->m_data = m->m_ext.ext_buf; 680 m->m_len = 0; 681 m->m_flags = (flags | M_EXT); 682 m->m_type = type; 683 684 if (flags & M_PKTHDR) { 685 m->m_pkthdr.rcvif = NULL; 686 m->m_pkthdr.len = 0; 687 m->m_pkthdr.header = NULL; 688 m->m_pkthdr.csum_flags = 0; 689 m->m_pkthdr.csum_data = 0; 690 m->m_pkthdr.tso_segsz = 0; 691 m->m_pkthdr.ether_vtag = 0; 692 m->m_pkthdr.flowid = 0; 693 SLIST_INIT(&m->m_pkthdr.tags); 694#ifdef MAC 695 /* If the label init fails, fail the alloc */ 696 error = mac_mbuf_init(m, how); 697 if (error) 698 return (error); 699#endif 700 } 701 /* m_ext is already initialized. */ 702 703 return (0); 704} 705 706int 707m_pkthdr_init(struct mbuf *m, int how) 708{ 709#ifdef MAC 710 int error; 711#endif 712 m->m_data = m->m_pktdat; 713 SLIST_INIT(&m->m_pkthdr.tags); 714 m->m_pkthdr.rcvif = NULL; 715 m->m_pkthdr.header = NULL; 716 m->m_pkthdr.len = 0; 717 m->m_pkthdr.flowid = 0; 718 m->m_pkthdr.csum_flags = 0; 719 m->m_pkthdr.csum_data = 0; 720 m->m_pkthdr.tso_segsz = 0; 721 m->m_pkthdr.ether_vtag = 0; 722#ifdef MAC 723 /* If the label init fails, fail the alloc */ 724 error = mac_mbuf_init(m, how); 725 if (error) 726 return (error); 727#endif 728 729 return (0); 730} 731 732/* 733 * This is the protocol drain routine. 734 * 735 * No locks should be held when this is called. The drain routines have to 736 * presently acquire some locks which raises the possibility of lock order 737 * reversal. 738 */ 739static void 740mb_reclaim(void *junk) 741{ 742 struct domain *dp; 743 struct protosw *pr; 744 745 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK | WARN_PANIC, NULL, 746 "mb_reclaim()"); 747 748 for (dp = domains; dp != NULL; dp = dp->dom_next) 749 for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) 750 if (pr->pr_drain != NULL) 751 (*pr->pr_drain)(); 752} 753