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