kern_mbuf.c revision 147537
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 147537 2005-06-23 04:33:39Z silby $"); 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 83int nmbclusters; 84struct mbstat mbstat; 85 86static void 87tunable_mbinit(void *dummy) 88{ 89 90 /* This has to be done before VM init. */ 91 nmbclusters = 1024 + maxusers * 64; 92 TUNABLE_INT_FETCH("kern.ipc.nmbclusters", &nmbclusters); 93} 94SYSINIT(tunable_mbinit, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_mbinit, NULL); 95 96SYSCTL_DECL(_kern_ipc); 97SYSCTL_INT(_kern_ipc, OID_AUTO, nmbclusters, CTLFLAG_RW, &nmbclusters, 0, 98 "Maximum number of mbuf clusters allowed"); 99SYSCTL_STRUCT(_kern_ipc, OID_AUTO, mbstat, CTLFLAG_RD, &mbstat, mbstat, 100 "Mbuf general information and statistics"); 101 102/* 103 * Zones from which we allocate. 104 */ 105uma_zone_t zone_mbuf; 106uma_zone_t zone_clust; 107uma_zone_t zone_pack; 108 109/* 110 * Local prototypes. 111 */ 112static int mb_ctor_mbuf(void *, int, void *, int); 113static int mb_ctor_clust(void *, int, void *, int); 114static int mb_ctor_pack(void *, int, void *, int); 115static void mb_dtor_mbuf(void *, int, void *); 116static void mb_dtor_clust(void *, int, void *); /* XXX */ 117static void mb_dtor_pack(void *, int, void *); /* XXX */ 118static int mb_init_pack(void *, int, int); 119static void mb_fini_pack(void *, int); 120 121static void mb_reclaim(void *); 122static void mbuf_init(void *); 123 124/* Ensure that MSIZE doesn't break dtom() - it must be a power of 2 */ 125CTASSERT((((MSIZE - 1) ^ MSIZE) + 1) >> 1 == MSIZE); 126 127/* 128 * Initialize FreeBSD Network buffer allocation. 129 */ 130SYSINIT(mbuf, SI_SUB_MBUF, SI_ORDER_FIRST, mbuf_init, NULL) 131static void 132mbuf_init(void *dummy) 133{ 134 135 /* 136 * Configure UMA zones for Mbufs, Clusters, and Packets. 137 */ 138 zone_mbuf = uma_zcreate("Mbuf", MSIZE, mb_ctor_mbuf, mb_dtor_mbuf, 139#ifdef INVARIANTS 140 trash_init, trash_fini, MSIZE - 1, UMA_ZONE_MAXBUCKET); 141#else 142 NULL, NULL, MSIZE - 1, UMA_ZONE_MAXBUCKET); 143#endif 144 zone_clust = uma_zcreate("MbufClust", MCLBYTES, mb_ctor_clust, 145#ifdef INVARIANTS 146 mb_dtor_clust, trash_init, trash_fini, UMA_ALIGN_PTR, UMA_ZONE_REFCNT); 147#else 148 mb_dtor_clust, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_REFCNT); 149#endif 150 if (nmbclusters > 0) 151 uma_zone_set_max(zone_clust, nmbclusters); 152 zone_pack = uma_zsecond_create("Packet", mb_ctor_pack, mb_dtor_pack, 153 mb_init_pack, mb_fini_pack, zone_mbuf); 154 155 /* uma_prealloc() goes here */ 156 157 /* 158 * Hook event handler for low-memory situation, used to 159 * drain protocols and push data back to the caches (UMA 160 * later pushes it back to VM). 161 */ 162 EVENTHANDLER_REGISTER(vm_lowmem, mb_reclaim, NULL, 163 EVENTHANDLER_PRI_FIRST); 164 165 /* 166 * [Re]set counters and local statistics knobs. 167 * XXX Some of these should go and be replaced, but UMA stat 168 * gathering needs to be revised. 169 */ 170 mbstat.m_mbufs = 0; 171 mbstat.m_mclusts = 0; 172 mbstat.m_drain = 0; 173 mbstat.m_msize = MSIZE; 174 mbstat.m_mclbytes = MCLBYTES; 175 mbstat.m_minclsize = MINCLSIZE; 176 mbstat.m_mlen = MLEN; 177 mbstat.m_mhlen = MHLEN; 178 mbstat.m_numtypes = MT_NTYPES; 179 180 mbstat.m_mcfail = mbstat.m_mpfail = 0; 181 mbstat.sf_iocnt = 0; 182 mbstat.sf_allocwait = mbstat.sf_allocfail = 0; 183} 184 185/* 186 * Constructor for Mbuf master zone. 187 * 188 * The 'arg' pointer points to a mb_args structure which 189 * contains call-specific information required to support the 190 * mbuf allocation API. 191 */ 192static int 193mb_ctor_mbuf(void *mem, int size, void *arg, int how) 194{ 195 struct mbuf *m; 196 struct mb_args *args; 197#ifdef MAC 198 int error; 199#endif 200 int flags; 201 short type; 202 203#ifdef INVARIANTS 204 trash_ctor(mem, size, arg, how); 205#endif 206 m = (struct mbuf *)mem; 207 args = (struct mb_args *)arg; 208 flags = args->flags; 209 type = args->type; 210 211 m->m_type = type; 212 m->m_next = NULL; 213 m->m_nextpkt = NULL; 214 m->m_flags = flags; 215 if (flags & M_PKTHDR) { 216 m->m_data = m->m_pktdat; 217 m->m_pkthdr.rcvif = NULL; 218 m->m_pkthdr.csum_flags = 0; 219 SLIST_INIT(&m->m_pkthdr.tags); 220#ifdef MAC 221 /* If the label init fails, fail the alloc */ 222 error = mac_init_mbuf(m, how); 223 if (error) 224 return (error); 225#endif 226 } else 227 m->m_data = m->m_dat; 228 mbstat.m_mbufs += 1; /* XXX */ 229 return (0); 230} 231 232/* 233 * The Mbuf master zone and Packet secondary zone destructor. 234 */ 235static void 236mb_dtor_mbuf(void *mem, int size, void *arg) 237{ 238 struct mbuf *m; 239 240 m = (struct mbuf *)mem; 241 if ((m->m_flags & M_PKTHDR) != 0) 242 m_tag_delete_chain(m, NULL); 243#ifdef INVARIANTS 244 trash_dtor(mem, size, arg); 245#endif 246 mbstat.m_mbufs -= 1; /* XXX */ 247} 248 249/* XXX Only because of stats */ 250static void 251mb_dtor_pack(void *mem, int size, void *arg) 252{ 253 struct mbuf *m; 254 255 m = (struct mbuf *)mem; 256 if ((m->m_flags & M_PKTHDR) != 0) 257 m_tag_delete_chain(m, NULL); 258#ifdef INVARIANTS 259 trash_dtor(m->m_ext.ext_buf, MCLBYTES, arg); 260#endif 261 mbstat.m_mbufs -= 1; /* XXX */ 262 mbstat.m_mclusts -= 1; /* XXX */ 263} 264 265/* 266 * The Cluster zone constructor. 267 * 268 * Here the 'arg' pointer points to the Mbuf which we 269 * are configuring cluster storage for. 270 */ 271static int 272mb_ctor_clust(void *mem, int size, void *arg, int how) 273{ 274 struct mbuf *m; 275 276#ifdef INVARIANTS 277 trash_ctor(mem, size, arg, how); 278#endif 279 m = (struct mbuf *)arg; 280 m->m_ext.ext_buf = (caddr_t)mem; 281 m->m_data = m->m_ext.ext_buf; 282 m->m_flags |= M_EXT; 283 m->m_ext.ext_free = NULL; 284 m->m_ext.ext_args = NULL; 285 m->m_ext.ext_size = MCLBYTES; 286 m->m_ext.ext_type = EXT_CLUSTER; 287 m->m_ext.ref_cnt = NULL; /* Lazy counter assign. */ 288 mbstat.m_mclusts += 1; /* XXX */ 289 return (0); 290} 291 292/* XXX */ 293static void 294mb_dtor_clust(void *mem, int size, void *arg) 295{ 296#ifdef INVARIANTS 297 trash_dtor(mem, size, arg); 298#endif 299 mbstat.m_mclusts -= 1; /* XXX */ 300} 301 302/* 303 * The Packet secondary zone's init routine, executed on the 304 * object's transition from keg slab to zone cache. 305 */ 306static int 307mb_init_pack(void *mem, int size, int how) 308{ 309 struct mbuf *m; 310 311 m = (struct mbuf *)mem; 312 m->m_ext.ext_buf = NULL; 313 uma_zalloc_arg(zone_clust, m, how); 314 if (m->m_ext.ext_buf == NULL) 315 return (ENOMEM); 316#ifdef INVARIANTS 317 trash_init(m->m_ext.ext_buf, MCLBYTES, how); 318#endif 319 mbstat.m_mclusts -= 1; /* XXX */ 320 return (0); 321} 322 323/* 324 * The Packet secondary zone's fini routine, executed on the 325 * object's transition from zone cache to keg slab. 326 */ 327static void 328mb_fini_pack(void *mem, int size) 329{ 330 struct mbuf *m; 331 332 m = (struct mbuf *)mem; 333#ifdef INVARIANTS 334 trash_fini(m->m_ext.ext_buf, MCLBYTES); 335#endif 336 uma_zfree_arg(zone_clust, m->m_ext.ext_buf, NULL); 337 m->m_ext.ext_buf = NULL; 338 mbstat.m_mclusts += 1; /* XXX */ 339} 340 341/* 342 * The "packet" keg constructor. 343 */ 344static int 345mb_ctor_pack(void *mem, int size, void *arg, int how) 346{ 347 struct mbuf *m; 348 struct mb_args *args; 349#ifdef MAC 350 int error; 351#endif 352 int flags; 353 short type; 354 355 m = (struct mbuf *)mem; 356 args = (struct mb_args *)arg; 357 flags = args->flags; 358 type = args->type; 359 360#ifdef INVARIANTS 361 trash_ctor(m->m_ext.ext_buf, MCLBYTES, arg, how); 362#endif 363 m->m_type = type; 364 m->m_next = NULL; 365 m->m_nextpkt = NULL; 366 m->m_data = m->m_ext.ext_buf; 367 m->m_flags = flags|M_EXT; 368 m->m_ext.ext_free = NULL; 369 m->m_ext.ext_args = NULL; 370 m->m_ext.ext_size = MCLBYTES; 371 m->m_ext.ext_type = EXT_PACKET; 372 m->m_ext.ref_cnt = NULL; /* Lazy counter assign. */ 373 374 if (flags & M_PKTHDR) { 375 m->m_pkthdr.rcvif = NULL; 376 m->m_pkthdr.csum_flags = 0; 377 SLIST_INIT(&m->m_pkthdr.tags); 378#ifdef MAC 379 /* If the label init fails, fail the alloc */ 380 error = mac_init_mbuf(m, how); 381 if (error) 382 return (error); 383#endif 384 } 385 mbstat.m_mbufs += 1; /* XXX */ 386 mbstat.m_mclusts += 1; /* XXX */ 387 return (0); 388} 389 390/* 391 * This is the protocol drain routine. 392 * 393 * No locks should be held when this is called. The drain routines have to 394 * presently acquire some locks which raises the possibility of lock order 395 * reversal. 396 */ 397static void 398mb_reclaim(void *junk) 399{ 400 struct domain *dp; 401 struct protosw *pr; 402 403 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK | WARN_PANIC, NULL, 404 "mb_reclaim()"); 405 406 mbstat.m_drain++; 407 for (dp = domains; dp != NULL; dp = dp->dom_next) 408 for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) 409 if (pr->pr_drain != NULL) 410 (*pr->pr_drain)(); 411} 412