1 /* 2 * Copyright (c) 2004 Jeffrey M. Hsu. All rights reserved. 3 * Copyright (c) 2004 The DragonFly Project. All rights reserved. 4 * 5 * This code is derived from software contributed to The DragonFly Project 6 * by Jeffrey M. Hsu. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. Neither the name of The DragonFly Project nor the names of its 17 * contributors may be used to endorse or promote products derived 18 * from this software without specific, prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 */ 33 34 /* 35 * Copyright (c) 2004 Jeffrey M. Hsu. All rights reserved. 36 * 37 * License terms: all terms for the DragonFly license above plus the following: 38 * 39 * 4. All advertising materials mentioning features or use of this software 40 * must display the following acknowledgement: 41 * 42 * This product includes software developed by Jeffrey M. Hsu 43 * for the DragonFly Project. 44 * 45 * This requirement may be waived with permission from Jeffrey Hsu. 46 * This requirement will sunset and may be removed on July 8 2005, 47 * after which the standard DragonFly license (as shown above) will 48 * apply. 49 */ 50 51 /* 52 * Copyright (c) 1982, 1986, 1988, 1991, 1993 53 * The Regents of the University of California. All rights reserved. 54 * 55 * Redistribution and use in source and binary forms, with or without 56 * modification, are permitted provided that the following conditions 57 * are met: 58 * 1. Redistributions of source code must retain the above copyright 59 * notice, this list of conditions and the following disclaimer. 60 * 2. Redistributions in binary form must reproduce the above copyright 61 * notice, this list of conditions and the following disclaimer in the 62 * documentation and/or other materials provided with the distribution. 63 * 3. All advertising materials mentioning features or use of this software 64 * must display the following acknowledgement: 65 * This product includes software developed by the University of 66 * California, Berkeley and its contributors. 67 * 4. Neither the name of the University nor the names of its contributors 68 * may be used to endorse or promote products derived from this software 69 * without specific prior written permission. 70 * 71 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 72 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 73 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 74 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 75 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 76 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 77 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 78 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 79 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 80 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 81 * SUCH DAMAGE. 82 * 83 * @(#)uipc_mbuf.c 8.2 (Berkeley) 1/4/94 84 * $FreeBSD: src/sys/kern/uipc_mbuf.c,v 1.51.2.24 2003/04/15 06:59:29 silby Exp $ 85 * $DragonFly: src/sys/kern/uipc_mbuf.c,v 1.29 2004/11/18 01:42:26 dillon Exp $ 86 */ 87 88 #include "opt_param.h" 89 #include "opt_mbuf_stress_test.h" 90 #include <sys/param.h> 91 #include <sys/systm.h> 92 #include <sys/malloc.h> 93 #include <sys/mbuf.h> 94 #include <sys/kernel.h> 95 #include <sys/sysctl.h> 96 #include <sys/domain.h> 97 #include <sys/protosw.h> 98 #include <sys/uio.h> 99 #include <sys/thread.h> 100 #include <sys/globaldata.h> 101 #include <sys/thread2.h> 102 103 #include <vm/vm.h> 104 #include <vm/vm_kern.h> 105 #include <vm/vm_extern.h> 106 107 #ifdef INVARIANTS 108 #include <machine/cpu.h> 109 #endif 110 111 /* 112 * mbuf cluster meta-data 113 */ 114 typedef struct mbcluster { 115 struct mbcluster *mcl_next; 116 int32_t mcl_magic; 117 int32_t mcl_refs; 118 void *mcl_data; 119 } *mbcluster_t; 120 121 typedef struct mbuf *mbuf_t; 122 123 #define MCL_MAGIC 0x6d62636c 124 125 static void mbinit (void *); 126 SYSINIT(mbuf, SI_SUB_MBUF, SI_ORDER_FIRST, mbinit, NULL) 127 128 static u_long mbtypes[MT_NTYPES]; 129 130 struct mbstat mbstat; 131 int max_linkhdr; 132 int max_protohdr; 133 int max_hdr; 134 int max_datalen; 135 int m_defragpackets; 136 int m_defragbytes; 137 int m_defraguseless; 138 int m_defragfailure; 139 #ifdef MBUF_STRESS_TEST 140 int m_defragrandomfailures; 141 #endif 142 143 int nmbclusters; 144 int nmbufs; 145 u_int m_mballoc_wid = 0; 146 u_int m_clalloc_wid = 0; 147 148 SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RW, 149 &max_linkhdr, 0, ""); 150 SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RW, 151 &max_protohdr, 0, ""); 152 SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RW, &max_hdr, 0, ""); 153 SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RW, 154 &max_datalen, 0, ""); 155 SYSCTL_INT(_kern_ipc, OID_AUTO, mbuf_wait, CTLFLAG_RW, 156 &mbuf_wait, 0, ""); 157 SYSCTL_STRUCT(_kern_ipc, KIPC_MBSTAT, mbstat, CTLFLAG_RW, &mbstat, mbstat, ""); 158 SYSCTL_OPAQUE(_kern_ipc, OID_AUTO, mbtypes, CTLFLAG_RD, mbtypes, 159 sizeof(mbtypes), "LU", ""); 160 SYSCTL_INT(_kern_ipc, KIPC_NMBCLUSTERS, nmbclusters, CTLFLAG_RW, 161 &nmbclusters, 0, "Maximum number of mbuf clusters available"); 162 SYSCTL_INT(_kern_ipc, OID_AUTO, nmbufs, CTLFLAG_RW, &nmbufs, 0, 163 "Maximum number of mbufs available"); 164 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD, 165 &m_defragpackets, 0, ""); 166 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD, 167 &m_defragbytes, 0, ""); 168 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD, 169 &m_defraguseless, 0, ""); 170 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD, 171 &m_defragfailure, 0, ""); 172 #ifdef MBUF_STRESS_TEST 173 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW, 174 &m_defragrandomfailures, 0, ""); 175 #endif 176 177 static int mcl_pool_count; 178 static int mcl_pool_max = 20; 179 static int mcl_free_max = 1000; 180 static int mbuf_free_max = 5000; 181 182 SYSCTL_INT(_kern_ipc, OID_AUTO, mcl_pool_max, CTLFLAG_RW, &mcl_pool_max, 0, 183 "Maximum number of mbufs+cluster in free list"); 184 SYSCTL_INT(_kern_ipc, OID_AUTO, mcl_pool_count, CTLFLAG_RD, &mcl_pool_count, 0, 185 "Current number of mbufs+cluster in free list"); 186 SYSCTL_INT(_kern_ipc, OID_AUTO, mcl_free_max, CTLFLAG_RW, &mcl_free_max, 0, 187 "Maximum number of clusters on the free list"); 188 SYSCTL_INT(_kern_ipc, OID_AUTO, mbuf_free_max, CTLFLAG_RW, &mbuf_free_max, 0, 189 "Maximum number of mbufs on the free list"); 190 191 static MALLOC_DEFINE(M_MBUF, "mbuf", "mbuf"); 192 static MALLOC_DEFINE(M_MBUFCL, "mbufcl", "mbufcl"); 193 194 static mbuf_t mmbfree; 195 static mbcluster_t mclfree; 196 static struct mbuf *mcl_pool; 197 198 static void m_reclaim (void); 199 static int m_mballoc(int nmb, int how); 200 static int m_clalloc(int ncl, int how); 201 static struct mbuf *m_mballoc_wait(int caller, int type); 202 static void m_mclref(void *arg); 203 static void m_mclfree(void *arg); 204 205 #ifndef NMBCLUSTERS 206 #define NMBCLUSTERS (512 + maxusers * 16) 207 #endif 208 #ifndef NMBUFS 209 #define NMBUFS (nmbclusters * 4) 210 #endif 211 212 /* 213 * Perform sanity checks of tunables declared above. 214 */ 215 static void 216 tunable_mbinit(void *dummy) 217 { 218 219 /* 220 * This has to be done before VM init. 221 */ 222 nmbclusters = NMBCLUSTERS; 223 TUNABLE_INT_FETCH("kern.ipc.nmbclusters", &nmbclusters); 224 nmbufs = NMBUFS; 225 TUNABLE_INT_FETCH("kern.ipc.nmbufs", &nmbufs); 226 /* Sanity checks */ 227 if (nmbufs < nmbclusters * 2) 228 nmbufs = nmbclusters * 2; 229 230 return; 231 } 232 SYSINIT(tunable_mbinit, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_mbinit, NULL); 233 234 /* "number of clusters of pages" */ 235 #define NCL_INIT 1 236 237 #define NMB_INIT 16 238 239 /* ARGSUSED*/ 240 static void 241 mbinit(void *dummy) 242 { 243 mmbfree = NULL; 244 mclfree = NULL; 245 mbstat.m_msize = MSIZE; 246 mbstat.m_mclbytes = MCLBYTES; 247 mbstat.m_minclsize = MINCLSIZE; 248 mbstat.m_mlen = MLEN; 249 mbstat.m_mhlen = MHLEN; 250 251 crit_enter(); 252 if (m_mballoc(NMB_INIT, MB_DONTWAIT) == 0) 253 goto bad; 254 #if MCLBYTES <= PAGE_SIZE 255 if (m_clalloc(NCL_INIT, MB_DONTWAIT) == 0) 256 goto bad; 257 #else 258 /* It's OK to call contigmalloc in this context. */ 259 if (m_clalloc(16, MB_WAIT) == 0) 260 goto bad; 261 #endif 262 crit_exit(); 263 return; 264 bad: 265 crit_exit(); 266 panic("mbinit"); 267 } 268 269 /* 270 * Allocate at least nmb mbufs and place on mbuf free list. 271 * Returns the number of mbufs successfully allocated, 0 if none. 272 * 273 * Must be called while in a critical section. 274 */ 275 static int 276 m_mballoc(int nmb, int how) 277 { 278 int i; 279 struct mbuf *m; 280 281 /* 282 * If we've hit the mbuf limit, stop allocating (or trying to) 283 * in order to avoid exhausting kernel memory entirely. 284 */ 285 if ((nmb + mbstat.m_mbufs) > nmbufs) 286 return (0); 287 288 /* 289 * Attempt to allocate the requested number of mbufs, terminate when 290 * the allocation fails but if blocking is allowed allocate at least 291 * one. 292 */ 293 for (i = 0; i < nmb; ++i) { 294 m = malloc(MSIZE, M_MBUF, M_NOWAIT|M_NULLOK|M_ZERO); 295 if (m == NULL) { 296 if (how == MB_WAIT) { 297 mbstat.m_wait++; 298 m = malloc(MSIZE, M_MBUF, 299 M_WAITOK|M_NULLOK|M_ZERO); 300 } 301 if (m == NULL) 302 break; 303 } 304 m->m_next = mmbfree; 305 mmbfree = m; 306 ++mbstat.m_mbufs; 307 ++mbtypes[MT_FREE]; 308 how = MB_DONTWAIT; 309 } 310 return(i); 311 } 312 313 /* 314 * Once mbuf memory has been exhausted and if the call to the allocation macros 315 * (or, in some cases, functions) is with MB_WAIT, then it is necessary to rely 316 * solely on reclaimed mbufs. Here we wait for an mbuf to be freed for a 317 * designated (mbuf_wait) time. 318 */ 319 static struct mbuf * 320 m_mballoc_wait(int caller, int type) 321 { 322 struct mbuf *m; 323 324 crit_enter(); 325 m_mballoc_wid++; 326 if ((tsleep(&m_mballoc_wid, 0, "mballc", mbuf_wait)) == EWOULDBLOCK) 327 m_mballoc_wid--; 328 crit_exit(); 329 330 /* 331 * Now that we (think) that we've got something, we will redo an 332 * MGET, but avoid getting into another instance of m_mballoc_wait() 333 * XXX: We retry to fetch _even_ if the sleep timed out. This is left 334 * this way, purposely, in the [unlikely] case that an mbuf was 335 * freed but the sleep was not awakened in time. 336 */ 337 m = NULL; 338 switch (caller) { 339 case MGET_C: 340 MGET(m, MB_DONTWAIT, type); 341 break; 342 case MGETHDR_C: 343 MGETHDR(m, MB_DONTWAIT, type); 344 break; 345 default: 346 panic("m_mballoc_wait: invalid caller (%d)", caller); 347 } 348 349 crit_enter(); 350 if (m != NULL) { /* We waited and got something... */ 351 mbstat.m_wait++; 352 /* Wake up another if we have more free. */ 353 if (mmbfree != NULL) 354 MMBWAKEUP(); 355 } 356 crit_exit(); 357 return (m); 358 } 359 360 #if MCLBYTES > PAGE_SIZE 361 static int i_want_my_mcl; 362 363 static void 364 kproc_mclalloc(void) 365 { 366 int status; 367 368 crit_enter(); 369 for (;;) { 370 tsleep(&i_want_my_mcl, 0, "mclalloc", 0); 371 372 while (i_want_my_mcl > 0) { 373 if (m_clalloc(1, MB_WAIT) == 0) 374 printf("m_clalloc failed even in thread context!\n"); 375 --i_want_my_mcl; 376 } 377 } 378 /* not reached */ 379 crit_exit(); 380 } 381 382 static struct thread *mclallocthread; 383 static struct kproc_desc mclalloc_kp = { 384 "mclalloc", 385 kproc_mclalloc, 386 &mclallocthread 387 }; 388 SYSINIT(mclallocthread, SI_SUB_KTHREAD_UPDATE, SI_ORDER_ANY, kproc_start, 389 &mclalloc_kp); 390 #endif 391 392 /* 393 * Allocate at least nmb mbuf clusters and place on mbuf free list. 394 * Returns the number of mbuf clusters successfully allocated, 0 if none. 395 * 396 * Must be called while in a critical section. 397 */ 398 static int 399 m_clalloc(int ncl, int how) 400 { 401 static int last_report; 402 mbcluster_t mcl; 403 void *data; 404 int i; 405 406 /* 407 * If we've hit the mbuf cluster limit, stop allocating (or trying to). 408 */ 409 if ((ncl + mbstat.m_clusters) > nmbclusters) 410 ncl = 0; 411 412 /* 413 * Attempt to allocate the requested number of mbuf clusters, 414 * terminate when the allocation fails but if blocking is allowed 415 * allocate at least one. 416 * 417 * We need to allocate two structures for each cluster... a 418 * ref counting / governing structure and the actual data. MCLBYTES 419 * should be a power of 2 which means that the slab allocator will 420 * return a buffer that does not cross a page boundary. 421 */ 422 for (i = 0; i < ncl; ++i) { 423 /* 424 * Meta structure 425 */ 426 mcl = malloc(sizeof(*mcl), M_MBUFCL, M_NOWAIT|M_NULLOK|M_ZERO); 427 if (mcl == NULL && how == MB_WAIT) { 428 mbstat.m_wait++; 429 mcl = malloc(sizeof(*mcl), 430 M_MBUFCL, M_WAITOK|M_NULLOK|M_ZERO); 431 } 432 433 /* 434 * Physically contiguous data buffer. 435 */ 436 #if MCLBYTES > PAGE_SIZE 437 if (how != MB_WAIT) { 438 i_want_my_mcl += ncl - i; 439 wakeup(&i_want_my_mcl); 440 mbstat.m_wait++; 441 data = NULL; 442 } else { 443 data = contigmalloc_map(MCLBYTES, M_MBUFCL, 444 M_WAITOK, 0ul, ~0ul, PAGE_SIZE, 0, kernel_map); 445 } 446 #else 447 data = malloc(MCLBYTES, M_MBUFCL, M_NOWAIT|M_NULLOK); 448 if (data == NULL) { 449 if (how == MB_WAIT) { 450 mbstat.m_wait++; 451 data = malloc(MCLBYTES, M_MBUFCL, 452 M_WAITOK|M_NULLOK); 453 } 454 } 455 #endif 456 if (data == NULL) { 457 free(mcl, M_MBUFCL); 458 break; 459 } 460 mcl->mcl_next = mclfree; 461 mcl->mcl_data = data; 462 mcl->mcl_magic = MCL_MAGIC; 463 mcl->mcl_refs = 0; 464 mclfree = mcl; 465 ++mbstat.m_clfree; 466 ++mbstat.m_clusters; 467 how = MB_DONTWAIT; 468 } 469 470 /* 471 * If we could not allocate any report failure no more often then 472 * once a second. 473 */ 474 if (i == 0) { 475 mbstat.m_drops++; 476 if (ticks < last_report || (ticks - last_report) >= hz) { 477 last_report = ticks; 478 printf("All mbuf clusters exhausted, please see tuning(7).\n"); 479 } 480 } 481 return (i); 482 } 483 484 /* 485 * Once cluster memory has been exhausted and the allocation is called with 486 * MB_WAIT, we rely on the mclfree pointers. If nothing is free, we will 487 * sleep for a designated amount of time (mbuf_wait) or until we're woken up 488 * due to sudden mcluster availability. 489 * 490 * Must be called while in a critical section. 491 */ 492 static void 493 m_clalloc_wait(void) 494 { 495 /* If in interrupt context, and INVARIANTS, maintain sanity and die. */ 496 KASSERT(mycpu->gd_intr_nesting_level == 0, 497 ("CLALLOC: CANNOT WAIT IN INTERRUPT")); 498 499 /* 500 * Sleep until something's available or until we expire. 501 */ 502 m_clalloc_wid++; 503 if ((tsleep(&m_clalloc_wid, 0, "mclalc", mbuf_wait)) == EWOULDBLOCK) 504 m_clalloc_wid--; 505 506 /* 507 * Try the allocation once more, and if we see mor then two 508 * free entries wake up others as well. 509 */ 510 m_clalloc(1, MB_WAIT); 511 if (mclfree && mclfree->mcl_next) { 512 MCLWAKEUP(); 513 } 514 } 515 516 /* 517 * Return the number of references to this mbuf's data. 0 is returned 518 * if the mbuf is not M_EXT, a reference count is returned if it is 519 * M_EXT|M_EXT_CLUSTER, and 99 is returned if it is a special M_EXT. 520 */ 521 int 522 m_sharecount(struct mbuf *m) 523 { 524 int count; 525 526 switch(m->m_flags & (M_EXT|M_EXT_CLUSTER)) { 527 case 0: 528 count = 0; 529 break; 530 case M_EXT: 531 count = 99; 532 break; 533 case M_EXT|M_EXT_CLUSTER: 534 count = ((mbcluster_t)m->m_ext.ext_arg)->mcl_refs; 535 break; 536 default: 537 panic("bad mbuf flags: %p", m); 538 count = 0; 539 } 540 return(count); 541 } 542 543 /* 544 * change mbuf to new type 545 */ 546 void 547 m_chtype(struct mbuf *m, int type) 548 { 549 crit_enter(); 550 --mbtypes[m->m_type]; 551 ++mbtypes[type]; 552 m->m_type = type; 553 crit_exit(); 554 } 555 556 /* 557 * When MGET fails, ask protocols to free space when short of memory, 558 * then re-attempt to allocate an mbuf. 559 */ 560 struct mbuf * 561 m_retry(int how, int t) 562 { 563 struct mbuf *m; 564 565 /* 566 * Must only do the reclaim if not in an interrupt context. 567 */ 568 if (how == MB_WAIT) { 569 KASSERT(mycpu->gd_intr_nesting_level == 0, 570 ("MBALLOC: CANNOT WAIT IN INTERRUPT")); 571 m_reclaim(); 572 } 573 574 /* 575 * Try to pull a new mbuf out of the cache, if the cache is empty 576 * try to allocate a new one and if that doesn't work we give up. 577 */ 578 crit_enter(); 579 if ((m = mmbfree) == NULL) { 580 m_mballoc(1, how); 581 if ((m = mmbfree) == NULL) { 582 static int last_report; 583 584 mbstat.m_drops++; 585 crit_exit(); 586 if (ticks < last_report || 587 (ticks - last_report) >= hz) { 588 last_report = ticks; 589 printf("All mbufs exhausted, please see tuning(7).\n"); 590 } 591 return (NULL); 592 } 593 } 594 595 /* 596 * Cache case, adjust globals before leaving the critical section 597 */ 598 mmbfree = m->m_next; 599 mbtypes[MT_FREE]--; 600 mbtypes[t]++; 601 mbstat.m_wait++; 602 crit_exit(); 603 604 m->m_type = t; 605 m->m_next = NULL; 606 m->m_nextpkt = NULL; 607 m->m_data = m->m_dat; 608 m->m_flags = 0; 609 return (m); 610 } 611 612 /* 613 * As above; retry an MGETHDR. 614 */ 615 struct mbuf * 616 m_retryhdr(int how, int t) 617 { 618 struct mbuf *m; 619 620 /* 621 * Must only do the reclaim if not in an interrupt context. 622 */ 623 if (how == MB_WAIT) { 624 KASSERT(mycpu->gd_intr_nesting_level == 0, 625 ("MBALLOC: CANNOT WAIT IN INTERRUPT")); 626 m_reclaim(); 627 } 628 629 /* 630 * Try to pull a new mbuf out of the cache, if the cache is empty 631 * try to allocate a new one and if that doesn't work we give up. 632 */ 633 crit_enter(); 634 if ((m = mmbfree) == NULL) { 635 m_mballoc(1, how); 636 if ((m = mmbfree) == NULL) { 637 static int last_report; 638 639 mbstat.m_drops++; 640 crit_exit(); 641 if (ticks < last_report || 642 (ticks - last_report) >= hz) { 643 last_report = ticks; 644 printf("All mbufs exhausted, please see tuning(7).\n"); 645 } 646 return (NULL); 647 } 648 } 649 650 /* 651 * Cache case, adjust globals before leaving the critical section 652 */ 653 mmbfree = m->m_next; 654 mbtypes[MT_FREE]--; 655 mbtypes[t]++; 656 mbstat.m_wait++; 657 crit_exit(); 658 659 m->m_type = t; 660 m->m_next = NULL; 661 m->m_nextpkt = NULL; 662 m->m_data = m->m_pktdat; 663 m->m_flags = M_PKTHDR; 664 m->m_pkthdr.rcvif = NULL; 665 SLIST_INIT(&m->m_pkthdr.tags); 666 m->m_pkthdr.csum_flags = 0; 667 return (m); 668 } 669 670 static void 671 m_reclaim(void) 672 { 673 struct domain *dp; 674 struct protosw *pr; 675 676 crit_enter(); 677 for (dp = domains; dp; dp = dp->dom_next) { 678 for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) { 679 if (pr->pr_drain) 680 (*pr->pr_drain)(); 681 } 682 } 683 crit_exit(); 684 mbstat.m_drain++; 685 } 686 687 /* 688 * Allocate an mbuf. If no mbufs are immediately available try to 689 * bring a bunch more into our cache (mmbfree list). A critical 690 * section is required to protect the mmbfree list and counters 691 * against interrupts. 692 */ 693 struct mbuf * 694 m_get(int how, int type) 695 { 696 struct mbuf *m; 697 698 /* 699 * Try to pull a new mbuf out of the cache, if the cache is empty 700 * try to allocate a new one and if that doesn't work try even harder 701 * by calling m_retryhdr(). 702 */ 703 crit_enter(); 704 if ((m = mmbfree) == NULL) { 705 m_mballoc(1, how); 706 if ((m = mmbfree) == NULL) { 707 crit_exit(); 708 m = m_retry(how, type); 709 if (m == NULL && how == MB_WAIT) 710 m = m_mballoc_wait(MGET_C, type); 711 return (m); 712 } 713 } 714 715 /* 716 * Cache case, adjust globals before leaving the critical section 717 */ 718 mmbfree = m->m_next; 719 mbtypes[MT_FREE]--; 720 mbtypes[type]++; 721 crit_exit(); 722 723 m->m_type = type; 724 m->m_next = NULL; 725 m->m_nextpkt = NULL; 726 m->m_data = m->m_dat; 727 m->m_flags = 0; 728 return (m); 729 } 730 731 struct mbuf * 732 m_gethdr(int how, int type) 733 { 734 struct mbuf *m; 735 736 /* 737 * Try to pull a new mbuf out of the cache, if the cache is empty 738 * try to allocate a new one and if that doesn't work try even harder 739 * by calling m_retryhdr(). 740 */ 741 crit_enter(); 742 if ((m = mmbfree) == NULL) { 743 m_mballoc(1, how); 744 if ((m = mmbfree) == NULL) { 745 crit_exit(); 746 m = m_retryhdr(how, type); 747 if (m == NULL && how == MB_WAIT) 748 m = m_mballoc_wait(MGETHDR_C, type); 749 return(m); 750 } 751 } 752 753 /* 754 * Cache case, adjust globals before leaving the critical section 755 */ 756 mmbfree = m->m_next; 757 mbtypes[MT_FREE]--; 758 mbtypes[type]++; 759 crit_exit(); 760 761 m->m_type = type; 762 m->m_next = NULL; 763 m->m_nextpkt = NULL; 764 m->m_data = m->m_pktdat; 765 m->m_flags = M_PKTHDR; 766 m->m_pkthdr.rcvif = NULL; 767 SLIST_INIT(&m->m_pkthdr.tags); 768 m->m_pkthdr.csum_flags = 0; 769 m->m_pkthdr.pf_flags = 0; 770 return (m); 771 } 772 773 struct mbuf * 774 m_getclr(int how, int type) 775 { 776 struct mbuf *m; 777 778 if ((m = m_get(how, type)) != NULL) { 779 bzero(mtod(m, caddr_t), MLEN); 780 } 781 return (m); 782 } 783 784 /* 785 * m_getcl() returns an mbuf with an attached cluster. 786 * Because many network drivers use this kind of buffers a lot, it is 787 * convenient to keep a small pool of free buffers of this kind. 788 * Even a small size such as 10 gives about 10% improvement in the 789 * forwarding rate in a bridge or router. 790 * The size of this free list is controlled by the sysctl variable 791 * mcl_pool_max. The list is populated on m_freem(), and used in 792 * m_getcl() if elements are available. 793 */ 794 struct mbuf * 795 m_getcl(int how, short type, int flags) 796 { 797 struct mbuf *mp; 798 799 crit_enter(); 800 if (flags & M_PKTHDR) { 801 if (type == MT_DATA && mcl_pool) { 802 mp = mcl_pool; 803 mcl_pool = mp->m_nextpkt; 804 --mcl_pool_count; 805 crit_exit(); 806 mp->m_nextpkt = NULL; 807 mp->m_data = mp->m_ext.ext_buf; 808 mp->m_flags = M_PKTHDR|M_EXT|M_EXT_CLUSTER; 809 mp->m_pkthdr.rcvif = NULL; 810 mp->m_pkthdr.csum_flags = 0; 811 return mp; 812 } 813 MGETHDR(mp, how, type); 814 } else { 815 MGET(mp, how, type); 816 } 817 if (mp) { 818 m_mclget(mp, how); 819 if ((mp->m_flags & M_EXT) == 0) { 820 m_free(mp); 821 mp = NULL; 822 } 823 } 824 crit_exit(); 825 return (mp); 826 } 827 828 /* 829 * struct mbuf * 830 * m_getm(m, len, how, type) 831 * 832 * This will allocate len-worth of mbufs and/or mbuf clusters (whatever fits 833 * best) and return a pointer to the top of the allocated chain. If m is 834 * non-null, then we assume that it is a single mbuf or an mbuf chain to 835 * which we want len bytes worth of mbufs and/or clusters attached, and so 836 * if we succeed in allocating it, we will just return a pointer to m. 837 * 838 * If we happen to fail at any point during the allocation, we will free 839 * up everything we have already allocated and return NULL. 840 * 841 */ 842 struct mbuf * 843 m_getm(struct mbuf *m, int len, int how, int type) 844 { 845 struct mbuf *top, *tail, *mp, *mtail = NULL; 846 847 KASSERT(len >= 0, ("len is < 0 in m_getm")); 848 849 mp = m_get(how, type); 850 if (mp == NULL) { 851 return (NULL); 852 } else if (len > MINCLSIZE) { 853 m_mclget(mp, how); 854 if ((mp->m_flags & M_EXT) == 0) { 855 m_free(mp); 856 return (NULL); 857 } 858 } 859 mp->m_len = 0; 860 len -= M_TRAILINGSPACE(mp); 861 862 if (m != NULL) { 863 for (mtail = m; mtail->m_next != NULL; mtail = mtail->m_next) 864 ; 865 } else { 866 m = mp; 867 } 868 869 top = tail = mp; 870 while (len > 0) { 871 mp = m_get(how, type); 872 if (mp == NULL) 873 goto failed; 874 875 tail->m_next = mp; 876 tail = mp; 877 if (len > MINCLSIZE) { 878 m_mclget(mp, how); 879 if ((mp->m_flags & M_EXT) == 0) 880 goto failed; 881 } 882 883 mp->m_len = 0; 884 len -= M_TRAILINGSPACE(mp); 885 } 886 887 if (mtail != NULL) 888 mtail->m_next = top; 889 return (m); 890 failed: 891 m_freem(top); 892 return (NULL); 893 } 894 895 /* 896 * m_mclget() - Adds a cluster to a normal mbuf, M_EXT is set on success. 897 */ 898 void 899 m_mclget(struct mbuf *m, int how) 900 { 901 mbcluster_t mcl; 902 903 KKASSERT((m->m_flags & M_EXT_OLD) == 0); 904 905 /* 906 * Allocate a cluster, return if we can't get one. 907 */ 908 crit_enter(); 909 if ((mcl = mclfree) == NULL) { 910 m_clalloc(1, how); 911 if ((mcl = mclfree) == NULL) { 912 if (how == MB_WAIT) { 913 m_clalloc_wait(); 914 mcl = mclfree; 915 } 916 if (mcl == NULL) { 917 crit_exit(); 918 return; 919 } 920 } 921 } 922 923 /* 924 * We have a cluster, unlink it from the free list and set the ref 925 * count. 926 */ 927 KKASSERT(mcl->mcl_refs == 0); 928 mclfree = mcl->mcl_next; 929 mcl->mcl_refs = 1; 930 --mbstat.m_clfree; 931 crit_exit(); 932 933 /* 934 * Add the cluster to the mbuf. The caller will detect that the 935 * mbuf now has an attached cluster. 936 */ 937 m->m_ext.ext_arg = mcl; 938 m->m_ext.ext_buf = mcl->mcl_data; 939 m->m_ext.ext_nref.new = m_mclref; 940 m->m_ext.ext_nfree.new = m_mclfree; 941 m->m_ext.ext_size = MCLBYTES; 942 943 m->m_data = m->m_ext.ext_buf; 944 m->m_flags |= M_EXT | M_EXT_CLUSTER; 945 } 946 947 static void 948 m_mclfree(void *arg) 949 { 950 mbcluster_t mcl = arg; 951 952 KKASSERT(mcl->mcl_magic == MCL_MAGIC); 953 KKASSERT(mcl->mcl_refs > 0); 954 crit_enter(); 955 if (--mcl->mcl_refs == 0) { 956 if (mbstat.m_clfree < mcl_free_max) { 957 mcl->mcl_next = mclfree; 958 mclfree = mcl; 959 ++mbstat.m_clfree; 960 MCLWAKEUP(); 961 } else { 962 mcl->mcl_magic = -1; 963 free(mcl->mcl_data, M_MBUFCL); 964 free(mcl, M_MBUFCL); 965 --mbstat.m_clusters; 966 } 967 } 968 crit_exit(); 969 } 970 971 static void 972 m_mclref(void *arg) 973 { 974 mbcluster_t mcl = arg; 975 976 KKASSERT(mcl->mcl_magic == MCL_MAGIC); 977 crit_enter(); 978 ++mcl->mcl_refs; 979 crit_exit(); 980 } 981 982 /* 983 * Helper routines for M_EXT reference/free 984 */ 985 static __inline void 986 m_extref(const struct mbuf *m) 987 { 988 KKASSERT(m->m_ext.ext_nfree.any != NULL); 989 crit_enter(); 990 if (m->m_flags & M_EXT_OLD) 991 m->m_ext.ext_nref.old(m->m_ext.ext_buf, m->m_ext.ext_size); 992 else 993 m->m_ext.ext_nref.new(m->m_ext.ext_arg); 994 crit_exit(); 995 } 996 997 /* 998 * m_free() 999 * 1000 * Free a single mbuf and any associated external storage. The successor, 1001 * if any, is returned. 1002 * 1003 * We do need to check non-first mbuf for m_aux, since some of existing 1004 * code does not call M_PREPEND properly. 1005 * (example: call to bpf_mtap from drivers) 1006 */ 1007 struct mbuf * 1008 m_free(struct mbuf *m) 1009 { 1010 struct mbuf *n; 1011 1012 crit_enter(); 1013 KASSERT(m->m_type != MT_FREE, ("freeing free mbuf %p", m)); 1014 1015 /* 1016 * Adjust our type count and delete any attached chains if the 1017 * mbuf is a packet header. 1018 */ 1019 if ((m->m_flags & M_PKTHDR) != 0) 1020 m_tag_delete_chain(m, NULL); 1021 1022 /* 1023 * Place the mbuf on the appropriate free list. Try to maintain a 1024 * small cache of mbuf+cluster pairs. 1025 */ 1026 n = m->m_next; 1027 m->m_next = NULL; 1028 if (m->m_flags & M_EXT) { 1029 KKASSERT(m->m_ext.ext_nfree.any != NULL); 1030 if (mcl_pool_count < mcl_pool_max && m && m->m_next == NULL && 1031 (m->m_flags & (M_PKTHDR|M_EXT_CLUSTER)) == (M_PKTHDR|M_EXT_CLUSTER) && 1032 m->m_type == MT_DATA && M_EXT_WRITABLE(m) ) { 1033 KKASSERT(((mbcluster_t)m->m_ext.ext_arg)->mcl_magic == MCL_MAGIC); 1034 m->m_nextpkt = mcl_pool; 1035 mcl_pool = m; 1036 ++mcl_pool_count; 1037 m = NULL; 1038 } else { 1039 if (m->m_flags & M_EXT_OLD) 1040 m->m_ext.ext_nfree.old(m->m_ext.ext_buf, m->m_ext.ext_size); 1041 else 1042 m->m_ext.ext_nfree.new(m->m_ext.ext_arg); 1043 m->m_flags = 0; 1044 m->m_ext.ext_arg = NULL; 1045 m->m_ext.ext_nref.new = NULL; 1046 m->m_ext.ext_nfree.new = NULL; 1047 } 1048 } 1049 if (m) { 1050 --mbtypes[m->m_type]; 1051 if (mbtypes[MT_FREE] < mbuf_free_max) { 1052 m->m_type = MT_FREE; 1053 mbtypes[MT_FREE]++; 1054 m->m_next = mmbfree; 1055 mmbfree = m; 1056 MMBWAKEUP(); 1057 } else { 1058 free(m, M_MBUF); 1059 --mbstat.m_mbufs; 1060 } 1061 } 1062 crit_exit(); 1063 return (n); 1064 } 1065 1066 void 1067 m_freem(struct mbuf *m) 1068 { 1069 crit_enter(); 1070 while (m) 1071 m = m_free(m); 1072 crit_exit(); 1073 } 1074 1075 /* 1076 * Mbuffer utility routines. 1077 */ 1078 1079 /* 1080 * Lesser-used path for M_PREPEND: 1081 * allocate new mbuf to prepend to chain, 1082 * copy junk along. 1083 */ 1084 struct mbuf * 1085 m_prepend(struct mbuf *m, int len, int how) 1086 { 1087 struct mbuf *mn; 1088 1089 MGET(mn, how, m->m_type); 1090 if (mn == (struct mbuf *)NULL) { 1091 m_freem(m); 1092 return ((struct mbuf *)NULL); 1093 } 1094 if (m->m_flags & M_PKTHDR) 1095 M_MOVE_PKTHDR(mn, m); 1096 mn->m_next = m; 1097 m = mn; 1098 if (len < MHLEN) 1099 MH_ALIGN(m, len); 1100 m->m_len = len; 1101 return (m); 1102 } 1103 1104 /* 1105 * Make a copy of an mbuf chain starting "off0" bytes from the beginning, 1106 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf. 1107 * The wait parameter is a choice of MB_WAIT/MB_DONTWAIT from caller. 1108 * Note that the copy is read-only, because clusters are not copied, 1109 * only their reference counts are incremented. 1110 */ 1111 #define MCFail (mbstat.m_mcfail) 1112 1113 struct mbuf * 1114 m_copym(const struct mbuf *m, int off0, int len, int wait) 1115 { 1116 struct mbuf *n, **np; 1117 int off = off0; 1118 struct mbuf *top; 1119 int copyhdr = 0; 1120 1121 KASSERT(off >= 0, ("m_copym, negative off %d", off)); 1122 KASSERT(len >= 0, ("m_copym, negative len %d", len)); 1123 if (off == 0 && m->m_flags & M_PKTHDR) 1124 copyhdr = 1; 1125 while (off > 0) { 1126 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain")); 1127 if (off < m->m_len) 1128 break; 1129 off -= m->m_len; 1130 m = m->m_next; 1131 } 1132 np = ⊤ 1133 top = 0; 1134 while (len > 0) { 1135 if (m == 0) { 1136 KASSERT(len == M_COPYALL, 1137 ("m_copym, length > size of mbuf chain")); 1138 break; 1139 } 1140 MGET(n, wait, m->m_type); 1141 *np = n; 1142 if (n == 0) 1143 goto nospace; 1144 if (copyhdr) { 1145 if (!m_dup_pkthdr(n, m, wait)) 1146 goto nospace; 1147 if (len == M_COPYALL) 1148 n->m_pkthdr.len -= off0; 1149 else 1150 n->m_pkthdr.len = len; 1151 copyhdr = 0; 1152 } 1153 n->m_len = min(len, m->m_len - off); 1154 if (m->m_flags & M_EXT) { 1155 n->m_data = m->m_data + off; 1156 m_extref(m); 1157 n->m_ext = m->m_ext; 1158 n->m_flags |= m->m_flags & 1159 (M_EXT | M_EXT_OLD | M_EXT_CLUSTER); 1160 } else { 1161 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t), 1162 (unsigned)n->m_len); 1163 } 1164 if (len != M_COPYALL) 1165 len -= n->m_len; 1166 off = 0; 1167 m = m->m_next; 1168 np = &n->m_next; 1169 } 1170 if (top == 0) 1171 MCFail++; 1172 return (top); 1173 nospace: 1174 m_freem(top); 1175 MCFail++; 1176 return (0); 1177 } 1178 1179 /* 1180 * Copy an entire packet, including header (which must be present). 1181 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'. 1182 * Note that the copy is read-only, because clusters are not copied, 1183 * only their reference counts are incremented. 1184 * Preserve alignment of the first mbuf so if the creator has left 1185 * some room at the beginning (e.g. for inserting protocol headers) 1186 * the copies also have the room available. 1187 */ 1188 struct mbuf * 1189 m_copypacket(struct mbuf *m, int how) 1190 { 1191 struct mbuf *top, *n, *o; 1192 1193 MGET(n, how, m->m_type); 1194 top = n; 1195 if (!n) 1196 goto nospace; 1197 1198 if (!m_dup_pkthdr(n, m, how)) 1199 goto nospace; 1200 n->m_len = m->m_len; 1201 if (m->m_flags & M_EXT) { 1202 n->m_data = m->m_data; 1203 m_extref(m); 1204 n->m_ext = m->m_ext; 1205 n->m_flags |= m->m_flags & (M_EXT | M_EXT_OLD | M_EXT_CLUSTER); 1206 } else { 1207 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat ); 1208 bcopy(mtod(m, char *), mtod(n, char *), n->m_len); 1209 } 1210 1211 m = m->m_next; 1212 while (m) { 1213 MGET(o, how, m->m_type); 1214 if (!o) 1215 goto nospace; 1216 1217 n->m_next = o; 1218 n = n->m_next; 1219 1220 n->m_len = m->m_len; 1221 if (m->m_flags & M_EXT) { 1222 n->m_data = m->m_data; 1223 m_extref(m); 1224 n->m_ext = m->m_ext; 1225 n->m_flags |= m->m_flags & 1226 (M_EXT | M_EXT_OLD | M_EXT_CLUSTER); 1227 } else { 1228 bcopy(mtod(m, char *), mtod(n, char *), n->m_len); 1229 } 1230 1231 m = m->m_next; 1232 } 1233 return top; 1234 nospace: 1235 m_freem(top); 1236 MCFail++; 1237 return 0; 1238 } 1239 1240 /* 1241 * Copy data from an mbuf chain starting "off" bytes from the beginning, 1242 * continuing for "len" bytes, into the indicated buffer. 1243 */ 1244 void 1245 m_copydata(const struct mbuf *m, int off, int len, caddr_t cp) 1246 { 1247 unsigned count; 1248 1249 KASSERT(off >= 0, ("m_copydata, negative off %d", off)); 1250 KASSERT(len >= 0, ("m_copydata, negative len %d", len)); 1251 while (off > 0) { 1252 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain")); 1253 if (off < m->m_len) 1254 break; 1255 off -= m->m_len; 1256 m = m->m_next; 1257 } 1258 while (len > 0) { 1259 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain")); 1260 count = min(m->m_len - off, len); 1261 bcopy(mtod(m, caddr_t) + off, cp, count); 1262 len -= count; 1263 cp += count; 1264 off = 0; 1265 m = m->m_next; 1266 } 1267 } 1268 1269 /* 1270 * Copy a packet header mbuf chain into a completely new chain, including 1271 * copying any mbuf clusters. Use this instead of m_copypacket() when 1272 * you need a writable copy of an mbuf chain. 1273 */ 1274 struct mbuf * 1275 m_dup(struct mbuf *m, int how) 1276 { 1277 struct mbuf **p, *top = NULL; 1278 int remain, moff, nsize; 1279 1280 /* Sanity check */ 1281 if (m == NULL) 1282 return (0); 1283 KASSERT((m->m_flags & M_PKTHDR) != 0, ("%s: !PKTHDR", __FUNCTION__)); 1284 1285 /* While there's more data, get a new mbuf, tack it on, and fill it */ 1286 remain = m->m_pkthdr.len; 1287 moff = 0; 1288 p = ⊤ 1289 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */ 1290 struct mbuf *n; 1291 1292 /* Get the next new mbuf */ 1293 MGET(n, how, m->m_type); 1294 if (n == NULL) 1295 goto nospace; 1296 if (top == NULL) { /* first one, must be PKTHDR */ 1297 if (!m_dup_pkthdr(n, m, how)) 1298 goto nospace; 1299 nsize = MHLEN; 1300 } else /* not the first one */ 1301 nsize = MLEN; 1302 if (remain >= MINCLSIZE) { 1303 MCLGET(n, how); 1304 if ((n->m_flags & M_EXT) == 0) { 1305 (void)m_free(n); 1306 goto nospace; 1307 } 1308 nsize = MCLBYTES; 1309 } 1310 n->m_len = 0; 1311 1312 /* Link it into the new chain */ 1313 *p = n; 1314 p = &n->m_next; 1315 1316 /* Copy data from original mbuf(s) into new mbuf */ 1317 while (n->m_len < nsize && m != NULL) { 1318 int chunk = min(nsize - n->m_len, m->m_len - moff); 1319 1320 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk); 1321 moff += chunk; 1322 n->m_len += chunk; 1323 remain -= chunk; 1324 if (moff == m->m_len) { 1325 m = m->m_next; 1326 moff = 0; 1327 } 1328 } 1329 1330 /* Check correct total mbuf length */ 1331 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL), 1332 ("%s: bogus m_pkthdr.len", __FUNCTION__)); 1333 } 1334 return (top); 1335 1336 nospace: 1337 m_freem(top); 1338 MCFail++; 1339 return (0); 1340 } 1341 1342 /* 1343 * Concatenate mbuf chain n to m. 1344 * Both chains must be of the same type (e.g. MT_DATA). 1345 * Any m_pkthdr is not updated. 1346 */ 1347 void 1348 m_cat(struct mbuf *m, struct mbuf *n) 1349 { 1350 while (m->m_next) 1351 m = m->m_next; 1352 while (n) { 1353 if (m->m_flags & M_EXT || 1354 m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) { 1355 /* just join the two chains */ 1356 m->m_next = n; 1357 return; 1358 } 1359 /* splat the data from one into the other */ 1360 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 1361 (u_int)n->m_len); 1362 m->m_len += n->m_len; 1363 n = m_free(n); 1364 } 1365 } 1366 1367 void 1368 m_adj(struct mbuf *mp, int req_len) 1369 { 1370 int len = req_len; 1371 struct mbuf *m; 1372 int count; 1373 1374 if ((m = mp) == NULL) 1375 return; 1376 if (len >= 0) { 1377 /* 1378 * Trim from head. 1379 */ 1380 while (m != NULL && len > 0) { 1381 if (m->m_len <= len) { 1382 len -= m->m_len; 1383 m->m_len = 0; 1384 m = m->m_next; 1385 } else { 1386 m->m_len -= len; 1387 m->m_data += len; 1388 len = 0; 1389 } 1390 } 1391 m = mp; 1392 if (mp->m_flags & M_PKTHDR) 1393 m->m_pkthdr.len -= (req_len - len); 1394 } else { 1395 /* 1396 * Trim from tail. Scan the mbuf chain, 1397 * calculating its length and finding the last mbuf. 1398 * If the adjustment only affects this mbuf, then just 1399 * adjust and return. Otherwise, rescan and truncate 1400 * after the remaining size. 1401 */ 1402 len = -len; 1403 count = 0; 1404 for (;;) { 1405 count += m->m_len; 1406 if (m->m_next == (struct mbuf *)0) 1407 break; 1408 m = m->m_next; 1409 } 1410 if (m->m_len >= len) { 1411 m->m_len -= len; 1412 if (mp->m_flags & M_PKTHDR) 1413 mp->m_pkthdr.len -= len; 1414 return; 1415 } 1416 count -= len; 1417 if (count < 0) 1418 count = 0; 1419 /* 1420 * Correct length for chain is "count". 1421 * Find the mbuf with last data, adjust its length, 1422 * and toss data from remaining mbufs on chain. 1423 */ 1424 m = mp; 1425 if (m->m_flags & M_PKTHDR) 1426 m->m_pkthdr.len = count; 1427 for (; m; m = m->m_next) { 1428 if (m->m_len >= count) { 1429 m->m_len = count; 1430 break; 1431 } 1432 count -= m->m_len; 1433 } 1434 while (m->m_next) 1435 (m = m->m_next) ->m_len = 0; 1436 } 1437 } 1438 1439 /* 1440 * Rearange an mbuf chain so that len bytes are contiguous 1441 * and in the data area of an mbuf (so that mtod will work for a structure 1442 * of size len). Returns the resulting mbuf chain on success, frees it and 1443 * returns null on failure. If there is room, it will add up to 1444 * max_protohdr-len extra bytes to the contiguous region in an attempt to 1445 * avoid being called next time. 1446 */ 1447 #define MPFail (mbstat.m_mpfail) 1448 1449 struct mbuf * 1450 m_pullup(struct mbuf *n, int len) 1451 { 1452 struct mbuf *m; 1453 int count; 1454 int space; 1455 1456 /* 1457 * If first mbuf has no cluster, and has room for len bytes 1458 * without shifting current data, pullup into it, 1459 * otherwise allocate a new mbuf to prepend to the chain. 1460 */ 1461 if ((n->m_flags & M_EXT) == 0 && 1462 n->m_data + len < &n->m_dat[MLEN] && n->m_next) { 1463 if (n->m_len >= len) 1464 return (n); 1465 m = n; 1466 n = n->m_next; 1467 len -= m->m_len; 1468 } else { 1469 if (len > MHLEN) 1470 goto bad; 1471 MGET(m, MB_DONTWAIT, n->m_type); 1472 if (m == 0) 1473 goto bad; 1474 m->m_len = 0; 1475 if (n->m_flags & M_PKTHDR) 1476 M_MOVE_PKTHDR(m, n); 1477 } 1478 space = &m->m_dat[MLEN] - (m->m_data + m->m_len); 1479 do { 1480 count = min(min(max(len, max_protohdr), space), n->m_len); 1481 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 1482 (unsigned)count); 1483 len -= count; 1484 m->m_len += count; 1485 n->m_len -= count; 1486 space -= count; 1487 if (n->m_len) 1488 n->m_data += count; 1489 else 1490 n = m_free(n); 1491 } while (len > 0 && n); 1492 if (len > 0) { 1493 (void) m_free(m); 1494 goto bad; 1495 } 1496 m->m_next = n; 1497 return (m); 1498 bad: 1499 m_freem(n); 1500 MPFail++; 1501 return (0); 1502 } 1503 1504 /* 1505 * Partition an mbuf chain in two pieces, returning the tail -- 1506 * all but the first len0 bytes. In case of failure, it returns NULL and 1507 * attempts to restore the chain to its original state. 1508 * 1509 * Note that the resulting mbufs might be read-only, because the new 1510 * mbuf can end up sharing an mbuf cluster with the original mbuf if 1511 * the "breaking point" happens to lie within a cluster mbuf. Use the 1512 * M_WRITABLE() macro to check for this case. 1513 */ 1514 struct mbuf * 1515 m_split(struct mbuf *m0, int len0, int wait) 1516 { 1517 struct mbuf *m, *n; 1518 unsigned len = len0, remain; 1519 1520 for (m = m0; m && len > m->m_len; m = m->m_next) 1521 len -= m->m_len; 1522 if (m == 0) 1523 return (0); 1524 remain = m->m_len - len; 1525 if (m0->m_flags & M_PKTHDR) { 1526 MGETHDR(n, wait, m0->m_type); 1527 if (n == 0) 1528 return (0); 1529 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif; 1530 n->m_pkthdr.len = m0->m_pkthdr.len - len0; 1531 m0->m_pkthdr.len = len0; 1532 if (m->m_flags & M_EXT) 1533 goto extpacket; 1534 if (remain > MHLEN) { 1535 /* m can't be the lead packet */ 1536 MH_ALIGN(n, 0); 1537 n->m_next = m_split(m, len, wait); 1538 if (n->m_next == 0) { 1539 (void) m_free(n); 1540 return (0); 1541 } else { 1542 n->m_len = 0; 1543 return (n); 1544 } 1545 } else 1546 MH_ALIGN(n, remain); 1547 } else if (remain == 0) { 1548 n = m->m_next; 1549 m->m_next = 0; 1550 return (n); 1551 } else { 1552 MGET(n, wait, m->m_type); 1553 if (n == 0) 1554 return (0); 1555 M_ALIGN(n, remain); 1556 } 1557 extpacket: 1558 if (m->m_flags & M_EXT) { 1559 n->m_data = m->m_data + len; 1560 m_extref(m); 1561 n->m_ext = m->m_ext; 1562 n->m_flags |= m->m_flags & (M_EXT | M_EXT_OLD | M_EXT_CLUSTER); 1563 } else { 1564 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain); 1565 } 1566 n->m_len = remain; 1567 m->m_len = len; 1568 n->m_next = m->m_next; 1569 m->m_next = 0; 1570 return (n); 1571 } 1572 /* 1573 * Routine to copy from device local memory into mbufs. 1574 */ 1575 struct mbuf * 1576 m_devget(char *buf, int totlen, int off0, struct ifnet *ifp, 1577 void (*copy) (char *from, caddr_t to, u_int len)) 1578 { 1579 struct mbuf *m; 1580 struct mbuf *top = 0, **mp = ⊤ 1581 int off = off0, len; 1582 char *cp; 1583 char *epkt; 1584 1585 cp = buf; 1586 epkt = cp + totlen; 1587 if (off) { 1588 cp += off + 2 * sizeof(u_short); 1589 totlen -= 2 * sizeof(u_short); 1590 } 1591 MGETHDR(m, MB_DONTWAIT, MT_DATA); 1592 if (m == 0) 1593 return (0); 1594 m->m_pkthdr.rcvif = ifp; 1595 m->m_pkthdr.len = totlen; 1596 m->m_len = MHLEN; 1597 1598 while (totlen > 0) { 1599 if (top) { 1600 MGET(m, MB_DONTWAIT, MT_DATA); 1601 if (m == 0) { 1602 m_freem(top); 1603 return (0); 1604 } 1605 m->m_len = MLEN; 1606 } 1607 len = min(totlen, epkt - cp); 1608 if (len >= MINCLSIZE) { 1609 MCLGET(m, MB_DONTWAIT); 1610 if (m->m_flags & M_EXT) 1611 m->m_len = len = min(len, MCLBYTES); 1612 else 1613 len = m->m_len; 1614 } else { 1615 /* 1616 * Place initial small packet/header at end of mbuf. 1617 */ 1618 if (len < m->m_len) { 1619 if (top == 0 && len + max_linkhdr <= m->m_len) 1620 m->m_data += max_linkhdr; 1621 m->m_len = len; 1622 } else 1623 len = m->m_len; 1624 } 1625 if (copy) 1626 copy(cp, mtod(m, caddr_t), (unsigned)len); 1627 else 1628 bcopy(cp, mtod(m, caddr_t), (unsigned)len); 1629 cp += len; 1630 *mp = m; 1631 mp = &m->m_next; 1632 totlen -= len; 1633 if (cp == epkt) 1634 cp = buf; 1635 } 1636 return (top); 1637 } 1638 1639 /* 1640 * Copy data from a buffer back into the indicated mbuf chain, 1641 * starting "off" bytes from the beginning, extending the mbuf 1642 * chain if necessary. 1643 */ 1644 void 1645 m_copyback(struct mbuf *m0, int off, int len, caddr_t cp) 1646 { 1647 int mlen; 1648 struct mbuf *m = m0, *n; 1649 int totlen = 0; 1650 1651 if (m0 == 0) 1652 return; 1653 while (off > (mlen = m->m_len)) { 1654 off -= mlen; 1655 totlen += mlen; 1656 if (m->m_next == 0) { 1657 n = m_getclr(MB_DONTWAIT, m->m_type); 1658 if (n == 0) 1659 goto out; 1660 n->m_len = min(MLEN, len + off); 1661 m->m_next = n; 1662 } 1663 m = m->m_next; 1664 } 1665 while (len > 0) { 1666 mlen = min (m->m_len - off, len); 1667 bcopy(cp, off + mtod(m, caddr_t), (unsigned)mlen); 1668 cp += mlen; 1669 len -= mlen; 1670 mlen += off; 1671 off = 0; 1672 totlen += mlen; 1673 if (len == 0) 1674 break; 1675 if (m->m_next == 0) { 1676 n = m_get(MB_DONTWAIT, m->m_type); 1677 if (n == 0) 1678 break; 1679 n->m_len = min(MLEN, len); 1680 m->m_next = n; 1681 } 1682 m = m->m_next; 1683 } 1684 out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen)) 1685 m->m_pkthdr.len = totlen; 1686 } 1687 1688 void 1689 m_print(const struct mbuf *m) 1690 { 1691 int len; 1692 const struct mbuf *m2; 1693 1694 len = m->m_pkthdr.len; 1695 m2 = m; 1696 while (len) { 1697 printf("%p %*D\n", m2, m2->m_len, (u_char *)m2->m_data, "-"); 1698 len -= m2->m_len; 1699 m2 = m2->m_next; 1700 } 1701 return; 1702 } 1703 1704 /* 1705 * "Move" mbuf pkthdr from "from" to "to". 1706 * "from" must have M_PKTHDR set, and "to" must be empty. 1707 */ 1708 void 1709 m_move_pkthdr(struct mbuf *to, struct mbuf *from) 1710 { 1711 KASSERT((to->m_flags & M_EXT) == 0, ("m_move_pkthdr: to has cluster")); 1712 1713 to->m_flags = from->m_flags & M_COPYFLAGS; 1714 to->m_data = to->m_pktdat; 1715 to->m_pkthdr = from->m_pkthdr; /* especially tags */ 1716 SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */ 1717 from->m_flags &= ~M_PKTHDR; 1718 } 1719 1720 /* 1721 * Duplicate "from"'s mbuf pkthdr in "to". 1722 * "from" must have M_PKTHDR set, and "to" must be empty. 1723 * In particular, this does a deep copy of the packet tags. 1724 */ 1725 int 1726 m_dup_pkthdr(struct mbuf *to, const struct mbuf *from, int how) 1727 { 1728 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT); 1729 if ((to->m_flags & M_EXT) == 0) 1730 to->m_data = to->m_pktdat; 1731 to->m_pkthdr = from->m_pkthdr; 1732 SLIST_INIT(&to->m_pkthdr.tags); 1733 return (m_tag_copy_chain(to, from, how)); 1734 } 1735 1736 /* 1737 * Defragment a mbuf chain, returning the shortest possible 1738 * chain of mbufs and clusters. If allocation fails and 1739 * this cannot be completed, NULL will be returned, but 1740 * the passed in chain will be unchanged. Upon success, 1741 * the original chain will be freed, and the new chain 1742 * will be returned. 1743 * 1744 * If a non-packet header is passed in, the original 1745 * mbuf (chain?) will be returned unharmed. 1746 */ 1747 struct mbuf * 1748 m_defrag(struct mbuf *m0, int how) 1749 { 1750 struct mbuf *m_new = NULL, *m_final = NULL; 1751 int progress = 0, length; 1752 1753 if (!(m0->m_flags & M_PKTHDR)) 1754 return (m0); 1755 1756 #ifdef MBUF_STRESS_TEST 1757 if (m_defragrandomfailures) { 1758 int temp = arc4random() & 0xff; 1759 if (temp == 0xba) 1760 goto nospace; 1761 } 1762 #endif 1763 1764 if (m0->m_pkthdr.len > MHLEN) 1765 m_final = m_getcl(how, MT_DATA, M_PKTHDR); 1766 else 1767 m_final = m_gethdr(how, MT_DATA); 1768 1769 if (m_final == NULL) 1770 goto nospace; 1771 1772 if (m_dup_pkthdr(m_final, m0, how) == NULL) 1773 goto nospace; 1774 1775 m_new = m_final; 1776 1777 while (progress < m0->m_pkthdr.len) { 1778 length = m0->m_pkthdr.len - progress; 1779 if (length > MCLBYTES) 1780 length = MCLBYTES; 1781 1782 if (m_new == NULL) { 1783 if (length > MLEN) 1784 m_new = m_getcl(how, MT_DATA, 0); 1785 else 1786 m_new = m_get(how, MT_DATA); 1787 if (m_new == NULL) 1788 goto nospace; 1789 } 1790 1791 m_copydata(m0, progress, length, mtod(m_new, caddr_t)); 1792 progress += length; 1793 m_new->m_len = length; 1794 if (m_new != m_final) 1795 m_cat(m_final, m_new); 1796 m_new = NULL; 1797 } 1798 if (m0->m_next == NULL) 1799 m_defraguseless++; 1800 m_freem(m0); 1801 m0 = m_final; 1802 m_defragpackets++; 1803 m_defragbytes += m0->m_pkthdr.len; 1804 return (m0); 1805 nospace: 1806 m_defragfailure++; 1807 if (m_new) 1808 m_free(m_new); 1809 if (m_final) 1810 m_freem(m_final); 1811 return (NULL); 1812 } 1813 1814 /* 1815 * Move data from uio into mbufs. 1816 * A length of zero means copy the whole uio. 1817 */ 1818 struct mbuf * 1819 m_uiomove(struct uio *uio, int wait, int len0) 1820 { 1821 struct mbuf *head; /* result mbuf chain */ 1822 struct mbuf *m; /* current working mbuf */ 1823 struct mbuf **mp; 1824 int resid, datalen, error; 1825 1826 resid = (len0 == 0) ? uio->uio_resid : min(len0, uio->uio_resid); 1827 1828 head = NULL; 1829 mp = &head; 1830 do { 1831 if (resid > MHLEN) { 1832 m = m_getcl(wait, MT_DATA, head == NULL ? M_PKTHDR : 0); 1833 if (m == NULL) 1834 goto failed; 1835 if (m->m_flags & M_PKTHDR) 1836 m->m_pkthdr.len = 0; 1837 } else { 1838 if (head == NULL) { 1839 MGETHDR(m, wait, MT_DATA); 1840 if (m == NULL) 1841 goto failed; 1842 m->m_pkthdr.len = 0; 1843 /* Leave room for protocol headers. */ 1844 if (resid < MHLEN) 1845 MH_ALIGN(m, resid); 1846 } else { 1847 MGET(m, wait, MT_DATA); 1848 if (m == NULL) 1849 goto failed; 1850 } 1851 } 1852 datalen = min(MCLBYTES, resid); 1853 error = uiomove(mtod(m, caddr_t), datalen, uio); 1854 if (error) { 1855 m_free(m); 1856 goto failed; 1857 } 1858 m->m_len = datalen; 1859 *mp = m; 1860 mp = &m->m_next; 1861 head->m_pkthdr.len += datalen; 1862 resid -= datalen; 1863 } while (resid > 0); 1864 1865 return (head); 1866 1867 failed: 1868 if (head) 1869 m_freem(head); 1870 return (NULL); 1871 } 1872