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 1.34 2005/02/20 00:20:43 joerg 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) { 428 if (how == MB_WAIT) { 429 mbstat.m_wait++; 430 mcl = malloc(sizeof(*mcl), 431 M_MBUFCL, M_WAITOK|M_NULLOK|M_ZERO); 432 } 433 if (mcl == NULL) 434 break; 435 } 436 437 /* 438 * Physically contiguous data buffer. 439 */ 440 #if MCLBYTES > PAGE_SIZE 441 if (how != MB_WAIT) { 442 i_want_my_mcl += ncl - i; 443 wakeup(&i_want_my_mcl); 444 mbstat.m_wait++; 445 data = NULL; 446 } else { 447 data = contigmalloc_map(MCLBYTES, M_MBUFCL, 448 M_WAITOK, 0ul, ~0ul, PAGE_SIZE, 0, kernel_map); 449 } 450 #else 451 data = malloc(MCLBYTES, M_MBUFCL, M_NOWAIT|M_NULLOK); 452 if (data == NULL) { 453 if (how == MB_WAIT) { 454 mbstat.m_wait++; 455 data = malloc(MCLBYTES, M_MBUFCL, 456 M_WAITOK|M_NULLOK); 457 } 458 } 459 #endif 460 if (data == NULL) { 461 free(mcl, M_MBUFCL); 462 break; 463 } 464 mcl->mcl_next = mclfree; 465 mcl->mcl_data = data; 466 mcl->mcl_magic = MCL_MAGIC; 467 mcl->mcl_refs = 0; 468 mclfree = mcl; 469 ++mbstat.m_clfree; 470 ++mbstat.m_clusters; 471 how = MB_DONTWAIT; 472 } 473 474 /* 475 * If we could not allocate any report failure no more often then 476 * once a second. 477 */ 478 if (i == 0) { 479 mbstat.m_drops++; 480 if (ticks < last_report || (ticks - last_report) >= hz) { 481 last_report = ticks; 482 printf("All mbuf clusters exhausted, please see tuning(7).\n"); 483 } 484 } 485 return (i); 486 } 487 488 /* 489 * Once cluster memory has been exhausted and the allocation is called with 490 * MB_WAIT, we rely on the mclfree pointers. If nothing is free, we will 491 * sleep for a designated amount of time (mbuf_wait) or until we're woken up 492 * due to sudden mcluster availability. 493 * 494 * Must be called while in a critical section. 495 */ 496 static void 497 m_clalloc_wait(void) 498 { 499 /* If in interrupt context, and INVARIANTS, maintain sanity and die. */ 500 KASSERT(mycpu->gd_intr_nesting_level == 0, 501 ("CLALLOC: CANNOT WAIT IN INTERRUPT")); 502 503 /* 504 * Sleep until something's available or until we expire. 505 */ 506 m_clalloc_wid++; 507 if ((tsleep(&m_clalloc_wid, 0, "mclalc", mbuf_wait)) == EWOULDBLOCK) 508 m_clalloc_wid--; 509 510 /* 511 * Try the allocation once more, and if we see mor then two 512 * free entries wake up others as well. 513 */ 514 m_clalloc(1, MB_WAIT); 515 if (mclfree && mclfree->mcl_next) { 516 MCLWAKEUP(); 517 } 518 } 519 520 /* 521 * Return the number of references to this mbuf's data. 0 is returned 522 * if the mbuf is not M_EXT, a reference count is returned if it is 523 * M_EXT|M_EXT_CLUSTER, and 99 is returned if it is a special M_EXT. 524 */ 525 int 526 m_sharecount(struct mbuf *m) 527 { 528 int count; 529 530 switch(m->m_flags & (M_EXT|M_EXT_CLUSTER)) { 531 case 0: 532 count = 0; 533 break; 534 case M_EXT: 535 count = 99; 536 break; 537 case M_EXT|M_EXT_CLUSTER: 538 count = ((mbcluster_t)m->m_ext.ext_arg)->mcl_refs; 539 break; 540 default: 541 panic("bad mbuf flags: %p", m); 542 count = 0; 543 } 544 return(count); 545 } 546 547 /* 548 * change mbuf to new type 549 */ 550 void 551 m_chtype(struct mbuf *m, int type) 552 { 553 crit_enter(); 554 --mbtypes[m->m_type]; 555 ++mbtypes[type]; 556 m->m_type = type; 557 crit_exit(); 558 } 559 560 /* 561 * When MGET fails, ask protocols to free space when short of memory, 562 * then re-attempt to allocate an mbuf. 563 */ 564 struct mbuf * 565 m_retry(int how, int t) 566 { 567 struct mbuf *m; 568 569 /* 570 * Must only do the reclaim if not in an interrupt context. 571 */ 572 if (how == MB_WAIT) { 573 KASSERT(mycpu->gd_intr_nesting_level == 0, 574 ("MBALLOC: CANNOT WAIT IN INTERRUPT")); 575 m_reclaim(); 576 } 577 578 /* 579 * Try to pull a new mbuf out of the cache, if the cache is empty 580 * try to allocate a new one and if that doesn't work we give up. 581 */ 582 crit_enter(); 583 if ((m = mmbfree) == NULL) { 584 m_mballoc(1, how); 585 if ((m = mmbfree) == NULL) { 586 static int last_report; 587 588 mbstat.m_drops++; 589 crit_exit(); 590 if (ticks < last_report || 591 (ticks - last_report) >= hz) { 592 last_report = ticks; 593 printf("All mbufs exhausted, please see tuning(7).\n"); 594 } 595 return (NULL); 596 } 597 } 598 599 /* 600 * Cache case, adjust globals before leaving the critical section 601 */ 602 mmbfree = m->m_next; 603 mbtypes[MT_FREE]--; 604 mbtypes[t]++; 605 mbstat.m_wait++; 606 crit_exit(); 607 608 m->m_type = t; 609 m->m_next = NULL; 610 m->m_nextpkt = NULL; 611 m->m_data = m->m_dat; 612 m->m_flags = 0; 613 return (m); 614 } 615 616 /* 617 * As above; retry an MGETHDR. 618 */ 619 struct mbuf * 620 m_retryhdr(int how, int t) 621 { 622 struct mbuf *m; 623 624 /* 625 * Must only do the reclaim if not in an interrupt context. 626 */ 627 if (how == MB_WAIT) { 628 KASSERT(mycpu->gd_intr_nesting_level == 0, 629 ("MBALLOC: CANNOT WAIT IN INTERRUPT")); 630 m_reclaim(); 631 } 632 633 /* 634 * Try to pull a new mbuf out of the cache, if the cache is empty 635 * try to allocate a new one and if that doesn't work we give up. 636 */ 637 crit_enter(); 638 if ((m = mmbfree) == NULL) { 639 m_mballoc(1, how); 640 if ((m = mmbfree) == NULL) { 641 static int last_report; 642 643 mbstat.m_drops++; 644 crit_exit(); 645 if (ticks < last_report || 646 (ticks - last_report) >= hz) { 647 last_report = ticks; 648 printf("All mbufs exhausted, please see tuning(7).\n"); 649 } 650 return (NULL); 651 } 652 } 653 654 /* 655 * Cache case, adjust globals before leaving the critical section 656 */ 657 mmbfree = m->m_next; 658 mbtypes[MT_FREE]--; 659 mbtypes[t]++; 660 mbstat.m_wait++; 661 crit_exit(); 662 663 m->m_type = t; 664 m->m_next = NULL; 665 m->m_nextpkt = NULL; 666 m->m_data = m->m_pktdat; 667 m->m_flags = M_PKTHDR; 668 m->m_pkthdr.rcvif = NULL; 669 SLIST_INIT(&m->m_pkthdr.tags); 670 m->m_pkthdr.csum_flags = 0; 671 return (m); 672 } 673 674 static void 675 m_reclaim(void) 676 { 677 struct domain *dp; 678 struct protosw *pr; 679 680 crit_enter(); 681 for (dp = domains; dp; dp = dp->dom_next) { 682 for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) { 683 if (pr->pr_drain) 684 (*pr->pr_drain)(); 685 } 686 } 687 crit_exit(); 688 mbstat.m_drain++; 689 } 690 691 /* 692 * Allocate an mbuf. If no mbufs are immediately available try to 693 * bring a bunch more into our cache (mmbfree list). A critical 694 * section is required to protect the mmbfree list and counters 695 * against interrupts. 696 */ 697 struct mbuf * 698 m_get(int how, int type) 699 { 700 struct mbuf *m; 701 702 /* 703 * Try to pull a new mbuf out of the cache, if the cache is empty 704 * try to allocate a new one and if that doesn't work try even harder 705 * by calling m_retryhdr(). 706 */ 707 crit_enter(); 708 if ((m = mmbfree) == NULL) { 709 m_mballoc(1, how); 710 if ((m = mmbfree) == NULL) { 711 crit_exit(); 712 m = m_retry(how, type); 713 if (m == NULL && how == MB_WAIT) 714 m = m_mballoc_wait(MGET_C, type); 715 return (m); 716 } 717 } 718 719 /* 720 * Cache case, adjust globals before leaving the critical section 721 */ 722 mmbfree = m->m_next; 723 mbtypes[MT_FREE]--; 724 mbtypes[type]++; 725 crit_exit(); 726 727 m->m_type = type; 728 m->m_next = NULL; 729 m->m_nextpkt = NULL; 730 m->m_data = m->m_dat; 731 m->m_flags = 0; 732 return (m); 733 } 734 735 struct mbuf * 736 m_gethdr(int how, int type) 737 { 738 struct mbuf *m; 739 740 /* 741 * Try to pull a new mbuf out of the cache, if the cache is empty 742 * try to allocate a new one and if that doesn't work try even harder 743 * by calling m_retryhdr(). 744 */ 745 crit_enter(); 746 if ((m = mmbfree) == NULL) { 747 m_mballoc(1, how); 748 if ((m = mmbfree) == NULL) { 749 crit_exit(); 750 m = m_retryhdr(how, type); 751 if (m == NULL && how == MB_WAIT) 752 m = m_mballoc_wait(MGETHDR_C, type); 753 return(m); 754 } 755 } 756 757 /* 758 * Cache case, adjust globals before leaving the critical section 759 */ 760 mmbfree = m->m_next; 761 mbtypes[MT_FREE]--; 762 mbtypes[type]++; 763 crit_exit(); 764 765 m->m_type = type; 766 m->m_next = NULL; 767 m->m_nextpkt = NULL; 768 m->m_data = m->m_pktdat; 769 m->m_flags = M_PKTHDR; 770 m->m_pkthdr.rcvif = NULL; 771 SLIST_INIT(&m->m_pkthdr.tags); 772 m->m_pkthdr.csum_flags = 0; 773 m->m_pkthdr.fw_flags = 0; 774 return (m); 775 } 776 777 struct mbuf * 778 m_getclr(int how, int type) 779 { 780 struct mbuf *m; 781 782 if ((m = m_get(how, type)) != NULL) { 783 bzero(mtod(m, caddr_t), MLEN); 784 } 785 return (m); 786 } 787 788 /* 789 * m_getcl() returns an mbuf with an attached cluster. 790 * Because many network drivers use this kind of buffers a lot, it is 791 * convenient to keep a small pool of free buffers of this kind. 792 * Even a small size such as 10 gives about 10% improvement in the 793 * forwarding rate in a bridge or router. 794 * The size of this free list is controlled by the sysctl variable 795 * mcl_pool_max. The list is populated on m_freem(), and used in 796 * m_getcl() if elements are available. 797 */ 798 struct mbuf * 799 m_getcl(int how, short type, int flags) 800 { 801 struct mbuf *mp; 802 803 crit_enter(); 804 if (flags & M_PKTHDR) { 805 if (type == MT_DATA && mcl_pool) { 806 mp = mcl_pool; 807 mcl_pool = mp->m_nextpkt; 808 --mcl_pool_count; 809 crit_exit(); 810 mp->m_nextpkt = NULL; 811 mp->m_data = mp->m_ext.ext_buf; 812 mp->m_flags = M_PKTHDR|M_EXT|M_EXT_CLUSTER; 813 mp->m_pkthdr.rcvif = NULL; 814 mp->m_pkthdr.csum_flags = 0; 815 return mp; 816 } 817 MGETHDR(mp, how, type); 818 } else { 819 MGET(mp, how, type); 820 } 821 if (mp) { 822 m_mclget(mp, how); 823 if ((mp->m_flags & M_EXT) == 0) { 824 m_free(mp); 825 mp = NULL; 826 } 827 } 828 crit_exit(); 829 return (mp); 830 } 831 832 /* 833 * struct mbuf * 834 * m_getm(m, len, how, type) 835 * 836 * This will allocate len-worth of mbufs and/or mbuf clusters (whatever fits 837 * best) and return a pointer to the top of the allocated chain. If m is 838 * non-null, then we assume that it is a single mbuf or an mbuf chain to 839 * which we want len bytes worth of mbufs and/or clusters attached, and so 840 * if we succeed in allocating it, we will just return a pointer to m. 841 * 842 * If we happen to fail at any point during the allocation, we will free 843 * up everything we have already allocated and return NULL. 844 * 845 */ 846 struct mbuf * 847 m_getm(struct mbuf *m, int len, int how, int type) 848 { 849 struct mbuf *top, *tail, *mp, *mtail = NULL; 850 851 KASSERT(len >= 0, ("len is < 0 in m_getm")); 852 853 mp = m_get(how, type); 854 if (mp == NULL) { 855 return (NULL); 856 } else if (len > MINCLSIZE) { 857 m_mclget(mp, how); 858 if ((mp->m_flags & M_EXT) == 0) { 859 m_free(mp); 860 return (NULL); 861 } 862 } 863 mp->m_len = 0; 864 len -= M_TRAILINGSPACE(mp); 865 866 if (m != NULL) { 867 for (mtail = m; mtail->m_next != NULL; mtail = mtail->m_next) 868 ; 869 } else { 870 m = mp; 871 } 872 873 top = tail = mp; 874 while (len > 0) { 875 mp = m_get(how, type); 876 if (mp == NULL) 877 goto failed; 878 879 tail->m_next = mp; 880 tail = mp; 881 if (len > MINCLSIZE) { 882 m_mclget(mp, how); 883 if ((mp->m_flags & M_EXT) == 0) 884 goto failed; 885 } 886 887 mp->m_len = 0; 888 len -= M_TRAILINGSPACE(mp); 889 } 890 891 if (mtail != NULL) 892 mtail->m_next = top; 893 return (m); 894 failed: 895 m_freem(top); 896 return (NULL); 897 } 898 899 /* 900 * m_mclget() - Adds a cluster to a normal mbuf, M_EXT is set on success. 901 */ 902 void 903 m_mclget(struct mbuf *m, int how) 904 { 905 mbcluster_t mcl; 906 907 KKASSERT((m->m_flags & M_EXT_OLD) == 0); 908 909 /* 910 * Allocate a cluster, return if we can't get one. 911 */ 912 crit_enter(); 913 if ((mcl = mclfree) == NULL) { 914 m_clalloc(1, how); 915 if ((mcl = mclfree) == NULL) { 916 if (how == MB_WAIT) { 917 m_clalloc_wait(); 918 mcl = mclfree; 919 } 920 if (mcl == NULL) { 921 crit_exit(); 922 return; 923 } 924 } 925 } 926 927 /* 928 * We have a cluster, unlink it from the free list and set the ref 929 * count. 930 */ 931 KKASSERT(mcl->mcl_refs == 0); 932 mclfree = mcl->mcl_next; 933 mcl->mcl_refs = 1; 934 --mbstat.m_clfree; 935 crit_exit(); 936 937 /* 938 * Add the cluster to the mbuf. The caller will detect that the 939 * mbuf now has an attached cluster. 940 */ 941 m->m_ext.ext_arg = mcl; 942 m->m_ext.ext_buf = mcl->mcl_data; 943 m->m_ext.ext_nref.new = m_mclref; 944 m->m_ext.ext_nfree.new = m_mclfree; 945 m->m_ext.ext_size = MCLBYTES; 946 947 m->m_data = m->m_ext.ext_buf; 948 m->m_flags |= M_EXT | M_EXT_CLUSTER; 949 } 950 951 static void 952 m_mclfree(void *arg) 953 { 954 mbcluster_t mcl = arg; 955 956 KKASSERT(mcl->mcl_magic == MCL_MAGIC); 957 KKASSERT(mcl->mcl_refs > 0); 958 crit_enter(); 959 if (--mcl->mcl_refs == 0) { 960 if (mbstat.m_clfree < mcl_free_max) { 961 mcl->mcl_next = mclfree; 962 mclfree = mcl; 963 ++mbstat.m_clfree; 964 MCLWAKEUP(); 965 } else { 966 mcl->mcl_magic = -1; 967 free(mcl->mcl_data, M_MBUFCL); 968 free(mcl, M_MBUFCL); 969 --mbstat.m_clusters; 970 } 971 } 972 crit_exit(); 973 } 974 975 static void 976 m_mclref(void *arg) 977 { 978 mbcluster_t mcl = arg; 979 980 KKASSERT(mcl->mcl_magic == MCL_MAGIC); 981 crit_enter(); 982 ++mcl->mcl_refs; 983 crit_exit(); 984 } 985 986 /* 987 * Helper routines for M_EXT reference/free 988 */ 989 static __inline void 990 m_extref(const struct mbuf *m) 991 { 992 KKASSERT(m->m_ext.ext_nfree.any != NULL); 993 crit_enter(); 994 if (m->m_flags & M_EXT_OLD) 995 m->m_ext.ext_nref.old(m->m_ext.ext_buf, m->m_ext.ext_size); 996 else 997 m->m_ext.ext_nref.new(m->m_ext.ext_arg); 998 crit_exit(); 999 } 1000 1001 /* 1002 * m_free() 1003 * 1004 * Free a single mbuf and any associated external storage. The successor, 1005 * if any, is returned. 1006 * 1007 * We do need to check non-first mbuf for m_aux, since some of existing 1008 * code does not call M_PREPEND properly. 1009 * (example: call to bpf_mtap from drivers) 1010 */ 1011 struct mbuf * 1012 m_free(struct mbuf *m) 1013 { 1014 struct mbuf *n; 1015 1016 crit_enter(); 1017 KASSERT(m->m_type != MT_FREE, ("freeing free mbuf %p", m)); 1018 1019 /* 1020 * Adjust our type count and delete any attached chains if the 1021 * mbuf is a packet header. 1022 */ 1023 if ((m->m_flags & M_PKTHDR) != 0) 1024 m_tag_delete_chain(m, NULL); 1025 1026 /* 1027 * Place the mbuf on the appropriate free list. Try to maintain a 1028 * small cache of mbuf+cluster pairs. 1029 */ 1030 n = m->m_next; 1031 m->m_next = NULL; 1032 if (m->m_flags & M_EXT) { 1033 KKASSERT(m->m_ext.ext_nfree.any != NULL); 1034 if (mcl_pool_count < mcl_pool_max && m && m->m_next == NULL && 1035 (m->m_flags & (M_PKTHDR|M_EXT_CLUSTER)) == (M_PKTHDR|M_EXT_CLUSTER) && 1036 m->m_type == MT_DATA && M_EXT_WRITABLE(m) ) { 1037 KKASSERT(((mbcluster_t)m->m_ext.ext_arg)->mcl_magic == MCL_MAGIC); 1038 m->m_nextpkt = mcl_pool; 1039 mcl_pool = m; 1040 ++mcl_pool_count; 1041 m = NULL; 1042 } else { 1043 if (m->m_flags & M_EXT_OLD) 1044 m->m_ext.ext_nfree.old(m->m_ext.ext_buf, m->m_ext.ext_size); 1045 else 1046 m->m_ext.ext_nfree.new(m->m_ext.ext_arg); 1047 m->m_flags = 0; 1048 m->m_ext.ext_arg = NULL; 1049 m->m_ext.ext_nref.new = NULL; 1050 m->m_ext.ext_nfree.new = NULL; 1051 } 1052 } 1053 if (m) { 1054 --mbtypes[m->m_type]; 1055 if (mbtypes[MT_FREE] < mbuf_free_max) { 1056 m->m_type = MT_FREE; 1057 mbtypes[MT_FREE]++; 1058 m->m_next = mmbfree; 1059 mmbfree = m; 1060 MMBWAKEUP(); 1061 } else { 1062 free(m, M_MBUF); 1063 --mbstat.m_mbufs; 1064 } 1065 } 1066 crit_exit(); 1067 return (n); 1068 } 1069 1070 void 1071 m_freem(struct mbuf *m) 1072 { 1073 crit_enter(); 1074 while (m) 1075 m = m_free(m); 1076 crit_exit(); 1077 } 1078 1079 /* 1080 * Mbuffer utility routines. 1081 */ 1082 1083 /* 1084 * Lesser-used path for M_PREPEND: 1085 * allocate new mbuf to prepend to chain, 1086 * copy junk along. 1087 */ 1088 struct mbuf * 1089 m_prepend(struct mbuf *m, int len, int how) 1090 { 1091 struct mbuf *mn; 1092 1093 MGET(mn, how, m->m_type); 1094 if (mn == (struct mbuf *)NULL) { 1095 m_freem(m); 1096 return ((struct mbuf *)NULL); 1097 } 1098 if (m->m_flags & M_PKTHDR) 1099 M_MOVE_PKTHDR(mn, m); 1100 mn->m_next = m; 1101 m = mn; 1102 if (len < MHLEN) 1103 MH_ALIGN(m, len); 1104 m->m_len = len; 1105 return (m); 1106 } 1107 1108 /* 1109 * Make a copy of an mbuf chain starting "off0" bytes from the beginning, 1110 * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf. 1111 * The wait parameter is a choice of MB_WAIT/MB_DONTWAIT from caller. 1112 * Note that the copy is read-only, because clusters are not copied, 1113 * only their reference counts are incremented. 1114 */ 1115 #define MCFail (mbstat.m_mcfail) 1116 1117 struct mbuf * 1118 m_copym(const struct mbuf *m, int off0, int len, int wait) 1119 { 1120 struct mbuf *n, **np; 1121 int off = off0; 1122 struct mbuf *top; 1123 int copyhdr = 0; 1124 1125 KASSERT(off >= 0, ("m_copym, negative off %d", off)); 1126 KASSERT(len >= 0, ("m_copym, negative len %d", len)); 1127 if (off == 0 && m->m_flags & M_PKTHDR) 1128 copyhdr = 1; 1129 while (off > 0) { 1130 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain")); 1131 if (off < m->m_len) 1132 break; 1133 off -= m->m_len; 1134 m = m->m_next; 1135 } 1136 np = ⊤ 1137 top = 0; 1138 while (len > 0) { 1139 if (m == 0) { 1140 KASSERT(len == M_COPYALL, 1141 ("m_copym, length > size of mbuf chain")); 1142 break; 1143 } 1144 MGET(n, wait, m->m_type); 1145 *np = n; 1146 if (n == 0) 1147 goto nospace; 1148 if (copyhdr) { 1149 if (!m_dup_pkthdr(n, m, wait)) 1150 goto nospace; 1151 if (len == M_COPYALL) 1152 n->m_pkthdr.len -= off0; 1153 else 1154 n->m_pkthdr.len = len; 1155 copyhdr = 0; 1156 } 1157 n->m_len = min(len, m->m_len - off); 1158 if (m->m_flags & M_EXT) { 1159 n->m_data = m->m_data + off; 1160 m_extref(m); 1161 n->m_ext = m->m_ext; 1162 n->m_flags |= m->m_flags & 1163 (M_EXT | M_EXT_OLD | M_EXT_CLUSTER); 1164 } else { 1165 bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t), 1166 (unsigned)n->m_len); 1167 } 1168 if (len != M_COPYALL) 1169 len -= n->m_len; 1170 off = 0; 1171 m = m->m_next; 1172 np = &n->m_next; 1173 } 1174 if (top == 0) 1175 MCFail++; 1176 return (top); 1177 nospace: 1178 m_freem(top); 1179 MCFail++; 1180 return (0); 1181 } 1182 1183 /* 1184 * Copy an entire packet, including header (which must be present). 1185 * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'. 1186 * Note that the copy is read-only, because clusters are not copied, 1187 * only their reference counts are incremented. 1188 * Preserve alignment of the first mbuf so if the creator has left 1189 * some room at the beginning (e.g. for inserting protocol headers) 1190 * the copies also have the room available. 1191 */ 1192 struct mbuf * 1193 m_copypacket(struct mbuf *m, int how) 1194 { 1195 struct mbuf *top, *n, *o; 1196 1197 MGET(n, how, m->m_type); 1198 top = n; 1199 if (!n) 1200 goto nospace; 1201 1202 if (!m_dup_pkthdr(n, m, how)) 1203 goto nospace; 1204 n->m_len = m->m_len; 1205 if (m->m_flags & M_EXT) { 1206 n->m_data = m->m_data; 1207 m_extref(m); 1208 n->m_ext = m->m_ext; 1209 n->m_flags |= m->m_flags & (M_EXT | M_EXT_OLD | M_EXT_CLUSTER); 1210 } else { 1211 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat ); 1212 bcopy(mtod(m, char *), mtod(n, char *), n->m_len); 1213 } 1214 1215 m = m->m_next; 1216 while (m) { 1217 MGET(o, how, m->m_type); 1218 if (!o) 1219 goto nospace; 1220 1221 n->m_next = o; 1222 n = n->m_next; 1223 1224 n->m_len = m->m_len; 1225 if (m->m_flags & M_EXT) { 1226 n->m_data = m->m_data; 1227 m_extref(m); 1228 n->m_ext = m->m_ext; 1229 n->m_flags |= m->m_flags & 1230 (M_EXT | M_EXT_OLD | M_EXT_CLUSTER); 1231 } else { 1232 bcopy(mtod(m, char *), mtod(n, char *), n->m_len); 1233 } 1234 1235 m = m->m_next; 1236 } 1237 return top; 1238 nospace: 1239 m_freem(top); 1240 MCFail++; 1241 return 0; 1242 } 1243 1244 /* 1245 * Copy data from an mbuf chain starting "off" bytes from the beginning, 1246 * continuing for "len" bytes, into the indicated buffer. 1247 */ 1248 void 1249 m_copydata(const struct mbuf *m, int off, int len, caddr_t cp) 1250 { 1251 unsigned count; 1252 1253 KASSERT(off >= 0, ("m_copydata, negative off %d", off)); 1254 KASSERT(len >= 0, ("m_copydata, negative len %d", len)); 1255 while (off > 0) { 1256 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain")); 1257 if (off < m->m_len) 1258 break; 1259 off -= m->m_len; 1260 m = m->m_next; 1261 } 1262 while (len > 0) { 1263 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain")); 1264 count = min(m->m_len - off, len); 1265 bcopy(mtod(m, caddr_t) + off, cp, count); 1266 len -= count; 1267 cp += count; 1268 off = 0; 1269 m = m->m_next; 1270 } 1271 } 1272 1273 /* 1274 * Copy a packet header mbuf chain into a completely new chain, including 1275 * copying any mbuf clusters. Use this instead of m_copypacket() when 1276 * you need a writable copy of an mbuf chain. 1277 */ 1278 struct mbuf * 1279 m_dup(struct mbuf *m, int how) 1280 { 1281 struct mbuf **p, *top = NULL; 1282 int remain, moff, nsize; 1283 1284 /* Sanity check */ 1285 if (m == NULL) 1286 return (0); 1287 KASSERT((m->m_flags & M_PKTHDR) != 0, ("%s: !PKTHDR", __func__)); 1288 1289 /* While there's more data, get a new mbuf, tack it on, and fill it */ 1290 remain = m->m_pkthdr.len; 1291 moff = 0; 1292 p = ⊤ 1293 while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */ 1294 struct mbuf *n; 1295 1296 /* Get the next new mbuf */ 1297 MGET(n, how, m->m_type); 1298 if (n == NULL) 1299 goto nospace; 1300 if (top == NULL) { /* first one, must be PKTHDR */ 1301 if (!m_dup_pkthdr(n, m, how)) 1302 goto nospace; 1303 nsize = MHLEN; 1304 } else /* not the first one */ 1305 nsize = MLEN; 1306 if (remain >= MINCLSIZE) { 1307 MCLGET(n, how); 1308 if ((n->m_flags & M_EXT) == 0) { 1309 (void)m_free(n); 1310 goto nospace; 1311 } 1312 nsize = MCLBYTES; 1313 } 1314 n->m_len = 0; 1315 1316 /* Link it into the new chain */ 1317 *p = n; 1318 p = &n->m_next; 1319 1320 /* Copy data from original mbuf(s) into new mbuf */ 1321 while (n->m_len < nsize && m != NULL) { 1322 int chunk = min(nsize - n->m_len, m->m_len - moff); 1323 1324 bcopy(m->m_data + moff, n->m_data + n->m_len, chunk); 1325 moff += chunk; 1326 n->m_len += chunk; 1327 remain -= chunk; 1328 if (moff == m->m_len) { 1329 m = m->m_next; 1330 moff = 0; 1331 } 1332 } 1333 1334 /* Check correct total mbuf length */ 1335 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL), 1336 ("%s: bogus m_pkthdr.len", __func__)); 1337 } 1338 return (top); 1339 1340 nospace: 1341 m_freem(top); 1342 MCFail++; 1343 return (0); 1344 } 1345 1346 /* 1347 * Concatenate mbuf chain n to m. 1348 * Both chains must be of the same type (e.g. MT_DATA). 1349 * Any m_pkthdr is not updated. 1350 */ 1351 void 1352 m_cat(struct mbuf *m, struct mbuf *n) 1353 { 1354 while (m->m_next) 1355 m = m->m_next; 1356 while (n) { 1357 if (m->m_flags & M_EXT || 1358 m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) { 1359 /* just join the two chains */ 1360 m->m_next = n; 1361 return; 1362 } 1363 /* splat the data from one into the other */ 1364 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 1365 (u_int)n->m_len); 1366 m->m_len += n->m_len; 1367 n = m_free(n); 1368 } 1369 } 1370 1371 void 1372 m_adj(struct mbuf *mp, int req_len) 1373 { 1374 int len = req_len; 1375 struct mbuf *m; 1376 int count; 1377 1378 if ((m = mp) == NULL) 1379 return; 1380 if (len >= 0) { 1381 /* 1382 * Trim from head. 1383 */ 1384 while (m != NULL && len > 0) { 1385 if (m->m_len <= len) { 1386 len -= m->m_len; 1387 m->m_len = 0; 1388 m = m->m_next; 1389 } else { 1390 m->m_len -= len; 1391 m->m_data += len; 1392 len = 0; 1393 } 1394 } 1395 m = mp; 1396 if (mp->m_flags & M_PKTHDR) 1397 m->m_pkthdr.len -= (req_len - len); 1398 } else { 1399 /* 1400 * Trim from tail. Scan the mbuf chain, 1401 * calculating its length and finding the last mbuf. 1402 * If the adjustment only affects this mbuf, then just 1403 * adjust and return. Otherwise, rescan and truncate 1404 * after the remaining size. 1405 */ 1406 len = -len; 1407 count = 0; 1408 for (;;) { 1409 count += m->m_len; 1410 if (m->m_next == (struct mbuf *)0) 1411 break; 1412 m = m->m_next; 1413 } 1414 if (m->m_len >= len) { 1415 m->m_len -= len; 1416 if (mp->m_flags & M_PKTHDR) 1417 mp->m_pkthdr.len -= len; 1418 return; 1419 } 1420 count -= len; 1421 if (count < 0) 1422 count = 0; 1423 /* 1424 * Correct length for chain is "count". 1425 * Find the mbuf with last data, adjust its length, 1426 * and toss data from remaining mbufs on chain. 1427 */ 1428 m = mp; 1429 if (m->m_flags & M_PKTHDR) 1430 m->m_pkthdr.len = count; 1431 for (; m; m = m->m_next) { 1432 if (m->m_len >= count) { 1433 m->m_len = count; 1434 break; 1435 } 1436 count -= m->m_len; 1437 } 1438 while (m->m_next) 1439 (m = m->m_next) ->m_len = 0; 1440 } 1441 } 1442 1443 /* 1444 * Rearange an mbuf chain so that len bytes are contiguous 1445 * and in the data area of an mbuf (so that mtod will work for a structure 1446 * of size len). Returns the resulting mbuf chain on success, frees it and 1447 * returns null on failure. If there is room, it will add up to 1448 * max_protohdr-len extra bytes to the contiguous region in an attempt to 1449 * avoid being called next time. 1450 */ 1451 #define MPFail (mbstat.m_mpfail) 1452 1453 struct mbuf * 1454 m_pullup(struct mbuf *n, int len) 1455 { 1456 struct mbuf *m; 1457 int count; 1458 int space; 1459 1460 /* 1461 * If first mbuf has no cluster, and has room for len bytes 1462 * without shifting current data, pullup into it, 1463 * otherwise allocate a new mbuf to prepend to the chain. 1464 */ 1465 if ((n->m_flags & M_EXT) == 0 && 1466 n->m_data + len < &n->m_dat[MLEN] && n->m_next) { 1467 if (n->m_len >= len) 1468 return (n); 1469 m = n; 1470 n = n->m_next; 1471 len -= m->m_len; 1472 } else { 1473 if (len > MHLEN) 1474 goto bad; 1475 MGET(m, MB_DONTWAIT, n->m_type); 1476 if (m == 0) 1477 goto bad; 1478 m->m_len = 0; 1479 if (n->m_flags & M_PKTHDR) 1480 M_MOVE_PKTHDR(m, n); 1481 } 1482 space = &m->m_dat[MLEN] - (m->m_data + m->m_len); 1483 do { 1484 count = min(min(max(len, max_protohdr), space), n->m_len); 1485 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 1486 (unsigned)count); 1487 len -= count; 1488 m->m_len += count; 1489 n->m_len -= count; 1490 space -= count; 1491 if (n->m_len) 1492 n->m_data += count; 1493 else 1494 n = m_free(n); 1495 } while (len > 0 && n); 1496 if (len > 0) { 1497 (void) m_free(m); 1498 goto bad; 1499 } 1500 m->m_next = n; 1501 return (m); 1502 bad: 1503 m_freem(n); 1504 MPFail++; 1505 return (0); 1506 } 1507 1508 /* 1509 * Partition an mbuf chain in two pieces, returning the tail -- 1510 * all but the first len0 bytes. In case of failure, it returns NULL and 1511 * attempts to restore the chain to its original state. 1512 * 1513 * Note that the resulting mbufs might be read-only, because the new 1514 * mbuf can end up sharing an mbuf cluster with the original mbuf if 1515 * the "breaking point" happens to lie within a cluster mbuf. Use the 1516 * M_WRITABLE() macro to check for this case. 1517 */ 1518 struct mbuf * 1519 m_split(struct mbuf *m0, int len0, int wait) 1520 { 1521 struct mbuf *m, *n; 1522 unsigned len = len0, remain; 1523 1524 for (m = m0; m && len > m->m_len; m = m->m_next) 1525 len -= m->m_len; 1526 if (m == 0) 1527 return (0); 1528 remain = m->m_len - len; 1529 if (m0->m_flags & M_PKTHDR) { 1530 MGETHDR(n, wait, m0->m_type); 1531 if (n == 0) 1532 return (0); 1533 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif; 1534 n->m_pkthdr.len = m0->m_pkthdr.len - len0; 1535 m0->m_pkthdr.len = len0; 1536 if (m->m_flags & M_EXT) 1537 goto extpacket; 1538 if (remain > MHLEN) { 1539 /* m can't be the lead packet */ 1540 MH_ALIGN(n, 0); 1541 n->m_next = m_split(m, len, wait); 1542 if (n->m_next == 0) { 1543 (void) m_free(n); 1544 return (0); 1545 } else { 1546 n->m_len = 0; 1547 return (n); 1548 } 1549 } else 1550 MH_ALIGN(n, remain); 1551 } else if (remain == 0) { 1552 n = m->m_next; 1553 m->m_next = 0; 1554 return (n); 1555 } else { 1556 MGET(n, wait, m->m_type); 1557 if (n == 0) 1558 return (0); 1559 M_ALIGN(n, remain); 1560 } 1561 extpacket: 1562 if (m->m_flags & M_EXT) { 1563 n->m_data = m->m_data + len; 1564 m_extref(m); 1565 n->m_ext = m->m_ext; 1566 n->m_flags |= m->m_flags & (M_EXT | M_EXT_OLD | M_EXT_CLUSTER); 1567 } else { 1568 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain); 1569 } 1570 n->m_len = remain; 1571 m->m_len = len; 1572 n->m_next = m->m_next; 1573 m->m_next = 0; 1574 return (n); 1575 } 1576 /* 1577 * Routine to copy from device local memory into mbufs. 1578 */ 1579 struct mbuf * 1580 m_devget(char *buf, int totlen, int off0, struct ifnet *ifp, 1581 void (*copy) (char *from, caddr_t to, u_int len)) 1582 { 1583 struct mbuf *m; 1584 struct mbuf *top = 0, **mp = ⊤ 1585 int off = off0, len; 1586 char *cp; 1587 char *epkt; 1588 1589 cp = buf; 1590 epkt = cp + totlen; 1591 if (off) { 1592 cp += off + 2 * sizeof(u_short); 1593 totlen -= 2 * sizeof(u_short); 1594 } 1595 MGETHDR(m, MB_DONTWAIT, MT_DATA); 1596 if (m == 0) 1597 return (0); 1598 m->m_pkthdr.rcvif = ifp; 1599 m->m_pkthdr.len = totlen; 1600 m->m_len = MHLEN; 1601 1602 while (totlen > 0) { 1603 if (top) { 1604 MGET(m, MB_DONTWAIT, MT_DATA); 1605 if (m == 0) { 1606 m_freem(top); 1607 return (0); 1608 } 1609 m->m_len = MLEN; 1610 } 1611 len = min(totlen, epkt - cp); 1612 if (len >= MINCLSIZE) { 1613 MCLGET(m, MB_DONTWAIT); 1614 if (m->m_flags & M_EXT) 1615 m->m_len = len = min(len, MCLBYTES); 1616 else 1617 len = m->m_len; 1618 } else { 1619 /* 1620 * Place initial small packet/header at end of mbuf. 1621 */ 1622 if (len < m->m_len) { 1623 if (top == 0 && len + max_linkhdr <= m->m_len) 1624 m->m_data += max_linkhdr; 1625 m->m_len = len; 1626 } else 1627 len = m->m_len; 1628 } 1629 if (copy) 1630 copy(cp, mtod(m, caddr_t), (unsigned)len); 1631 else 1632 bcopy(cp, mtod(m, caddr_t), (unsigned)len); 1633 cp += len; 1634 *mp = m; 1635 mp = &m->m_next; 1636 totlen -= len; 1637 if (cp == epkt) 1638 cp = buf; 1639 } 1640 return (top); 1641 } 1642 1643 /* 1644 * Copy data from a buffer back into the indicated mbuf chain, 1645 * starting "off" bytes from the beginning, extending the mbuf 1646 * chain if necessary. 1647 */ 1648 void 1649 m_copyback(struct mbuf *m0, int off, int len, caddr_t cp) 1650 { 1651 int mlen; 1652 struct mbuf *m = m0, *n; 1653 int totlen = 0; 1654 1655 if (m0 == 0) 1656 return; 1657 while (off > (mlen = m->m_len)) { 1658 off -= mlen; 1659 totlen += mlen; 1660 if (m->m_next == 0) { 1661 n = m_getclr(MB_DONTWAIT, m->m_type); 1662 if (n == 0) 1663 goto out; 1664 n->m_len = min(MLEN, len + off); 1665 m->m_next = n; 1666 } 1667 m = m->m_next; 1668 } 1669 while (len > 0) { 1670 mlen = min (m->m_len - off, len); 1671 bcopy(cp, off + mtod(m, caddr_t), (unsigned)mlen); 1672 cp += mlen; 1673 len -= mlen; 1674 mlen += off; 1675 off = 0; 1676 totlen += mlen; 1677 if (len == 0) 1678 break; 1679 if (m->m_next == 0) { 1680 n = m_get(MB_DONTWAIT, m->m_type); 1681 if (n == 0) 1682 break; 1683 n->m_len = min(MLEN, len); 1684 m->m_next = n; 1685 } 1686 m = m->m_next; 1687 } 1688 out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen)) 1689 m->m_pkthdr.len = totlen; 1690 } 1691 1692 void 1693 m_print(const struct mbuf *m) 1694 { 1695 int len; 1696 const struct mbuf *m2; 1697 1698 len = m->m_pkthdr.len; 1699 m2 = m; 1700 while (len) { 1701 printf("%p %*D\n", m2, m2->m_len, (u_char *)m2->m_data, "-"); 1702 len -= m2->m_len; 1703 m2 = m2->m_next; 1704 } 1705 return; 1706 } 1707 1708 /* 1709 * "Move" mbuf pkthdr from "from" to "to". 1710 * "from" must have M_PKTHDR set, and "to" must be empty. 1711 */ 1712 void 1713 m_move_pkthdr(struct mbuf *to, struct mbuf *from) 1714 { 1715 KASSERT((to->m_flags & M_EXT) == 0, ("m_move_pkthdr: to has cluster")); 1716 1717 to->m_flags = from->m_flags & M_COPYFLAGS; 1718 to->m_data = to->m_pktdat; 1719 to->m_pkthdr = from->m_pkthdr; /* especially tags */ 1720 SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */ 1721 from->m_flags &= ~M_PKTHDR; 1722 } 1723 1724 /* 1725 * Duplicate "from"'s mbuf pkthdr in "to". 1726 * "from" must have M_PKTHDR set, and "to" must be empty. 1727 * In particular, this does a deep copy of the packet tags. 1728 */ 1729 int 1730 m_dup_pkthdr(struct mbuf *to, const struct mbuf *from, int how) 1731 { 1732 to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT); 1733 if ((to->m_flags & M_EXT) == 0) 1734 to->m_data = to->m_pktdat; 1735 to->m_pkthdr = from->m_pkthdr; 1736 SLIST_INIT(&to->m_pkthdr.tags); 1737 return (m_tag_copy_chain(to, from, how)); 1738 } 1739 1740 /* 1741 * Defragment a mbuf chain, returning the shortest possible 1742 * chain of mbufs and clusters. If allocation fails and 1743 * this cannot be completed, NULL will be returned, but 1744 * the passed in chain will be unchanged. Upon success, 1745 * the original chain will be freed, and the new chain 1746 * will be returned. 1747 * 1748 * If a non-packet header is passed in, the original 1749 * mbuf (chain?) will be returned unharmed. 1750 * 1751 * m_defrag_nofree doesn't free the passed in mbuf. 1752 */ 1753 struct mbuf * 1754 m_defrag(struct mbuf *m0, int how) 1755 { 1756 struct mbuf *m_new; 1757 1758 if ((m_new = m_defrag_nofree(m0, how)) == NULL) 1759 return (NULL); 1760 if (m_new != m0) 1761 m_freem(m0); 1762 return (m_new); 1763 } 1764 1765 struct mbuf * 1766 m_defrag_nofree(struct mbuf *m0, int how) 1767 { 1768 struct mbuf *m_new = NULL, *m_final = NULL; 1769 int progress = 0, length; 1770 1771 if (!(m0->m_flags & M_PKTHDR)) 1772 return (m0); 1773 1774 #ifdef MBUF_STRESS_TEST 1775 if (m_defragrandomfailures) { 1776 int temp = arc4random() & 0xff; 1777 if (temp == 0xba) 1778 goto nospace; 1779 } 1780 #endif 1781 1782 if (m0->m_pkthdr.len > MHLEN) 1783 m_final = m_getcl(how, MT_DATA, M_PKTHDR); 1784 else 1785 m_final = m_gethdr(how, MT_DATA); 1786 1787 if (m_final == NULL) 1788 goto nospace; 1789 1790 if (m_dup_pkthdr(m_final, m0, how) == NULL) 1791 goto nospace; 1792 1793 m_new = m_final; 1794 1795 while (progress < m0->m_pkthdr.len) { 1796 length = m0->m_pkthdr.len - progress; 1797 if (length > MCLBYTES) 1798 length = MCLBYTES; 1799 1800 if (m_new == NULL) { 1801 if (length > MLEN) 1802 m_new = m_getcl(how, MT_DATA, 0); 1803 else 1804 m_new = m_get(how, MT_DATA); 1805 if (m_new == NULL) 1806 goto nospace; 1807 } 1808 1809 m_copydata(m0, progress, length, mtod(m_new, caddr_t)); 1810 progress += length; 1811 m_new->m_len = length; 1812 if (m_new != m_final) 1813 m_cat(m_final, m_new); 1814 m_new = NULL; 1815 } 1816 if (m0->m_next == NULL) 1817 m_defraguseless++; 1818 m_defragpackets++; 1819 m_defragbytes += m_final->m_pkthdr.len; 1820 return (m_final); 1821 nospace: 1822 m_defragfailure++; 1823 if (m_new) 1824 m_free(m_new); 1825 if (m_final) 1826 m_freem(m_final); 1827 return (NULL); 1828 } 1829 1830 /* 1831 * Move data from uio into mbufs. 1832 * A length of zero means copy the whole uio. 1833 */ 1834 struct mbuf * 1835 m_uiomove(struct uio *uio, int wait, int len0) 1836 { 1837 struct mbuf *head; /* result mbuf chain */ 1838 struct mbuf *m; /* current working mbuf */ 1839 struct mbuf **mp; 1840 int resid, datalen, error; 1841 1842 resid = (len0 == 0) ? uio->uio_resid : min(len0, uio->uio_resid); 1843 1844 head = NULL; 1845 mp = &head; 1846 do { 1847 if (resid > MHLEN) { 1848 m = m_getcl(wait, MT_DATA, head == NULL ? M_PKTHDR : 0); 1849 if (m == NULL) 1850 goto failed; 1851 if (m->m_flags & M_PKTHDR) 1852 m->m_pkthdr.len = 0; 1853 } else { 1854 if (head == NULL) { 1855 MGETHDR(m, wait, MT_DATA); 1856 if (m == NULL) 1857 goto failed; 1858 m->m_pkthdr.len = 0; 1859 /* Leave room for protocol headers. */ 1860 if (resid < MHLEN) 1861 MH_ALIGN(m, resid); 1862 } else { 1863 MGET(m, wait, MT_DATA); 1864 if (m == NULL) 1865 goto failed; 1866 } 1867 } 1868 datalen = min(MCLBYTES, resid); 1869 error = uiomove(mtod(m, caddr_t), datalen, uio); 1870 if (error) { 1871 m_free(m); 1872 goto failed; 1873 } 1874 m->m_len = datalen; 1875 *mp = m; 1876 mp = &m->m_next; 1877 head->m_pkthdr.len += datalen; 1878 resid -= datalen; 1879 } while (resid > 0); 1880 1881 return (head); 1882 1883 failed: 1884 if (head) 1885 m_freem(head); 1886 return (NULL); 1887 } 1888