1/*-
2 * Copyright (c) 2004, 2005,
3 * Bosko Milekic <bmilekic@FreeBSD.org>. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice unmodified, this list of conditions and the following
10 * disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 *
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25 * SUCH DAMAGE.
26 */
27
28#include <sys/cdefs.h>
29__FBSDID("$FreeBSD: head/sys/kern/kern_mbuf.c 243995 2012-12-07 22:19:41Z pjd $");
30
31#include "opt_param.h"
32
33#include <sys/param.h>
34#include <sys/malloc.h>
35#include <sys/systm.h>
36#include <sys/mbuf.h>
37#include <sys/domain.h>
38#include <sys/eventhandler.h>
39#include <sys/kernel.h>
40#include <sys/protosw.h>
41#include <sys/smp.h>
42#include <sys/sysctl.h>
43
44#include <security/mac/mac_framework.h>
45
46#include <vm/vm.h>
47#include <vm/vm_extern.h>
48#include <vm/vm_kern.h>
49#include <vm/vm_page.h>
50#include <vm/uma.h>
51#include <vm/uma_int.h>
52#include <vm/uma_dbg.h>
53
54/*
55 * In FreeBSD, Mbufs and Mbuf Clusters are allocated from UMA
56 * Zones.
57 *
58 * Mbuf Clusters (2K, contiguous) are allocated from the Cluster
59 * Zone. The Zone can be capped at kern.ipc.nmbclusters, if the
60 * administrator so desires.
61 *
62 * Mbufs are allocated from a UMA Master Zone called the Mbuf
63 * Zone.
64 *
65 * Additionally, FreeBSD provides a Packet Zone, which it
66 * configures as a Secondary Zone to the Mbuf Master Zone,
67 * thus sharing backend Slab kegs with the Mbuf Master Zone.
68 *
69 * Thus common-case allocations and locking are simplified:
70 *
71 * m_clget() m_getcl()
72 * | |
73 * | .------------>[(Packet Cache)] m_get(), m_gethdr()
74 * | | [ Packet ] |
75 * [(Cluster Cache)] [ Secondary ] [ (Mbuf Cache) ]
76 * [ Cluster Zone ] [ Zone ] [ Mbuf Master Zone ]
77 * | \________ |
78 * [ Cluster Keg ] \ /
79 * | [ Mbuf Keg ]
80 * [ Cluster Slabs ] |
81 * | [ Mbuf Slabs ]
82 * \____________(VM)_________________/
83 *
84 *
85 * Whenever an object is allocated with uma_zalloc() out of
86 * one of the Zones its _ctor_ function is executed. The same
87 * for any deallocation through uma_zfree() the _dtor_ function
88 * is executed.
89 *
90 * Caches are per-CPU and are filled from the Master Zone.
91 *
92 * Whenever an object is allocated from the underlying global
93 * memory pool it gets pre-initialized with the _zinit_ functions.
94 * When the Keg's are overfull objects get decomissioned with
95 * _zfini_ functions and free'd back to the global memory pool.
96 *
97 */
98
99int nmbufs; /* limits number of mbufs */
100int nmbclusters; /* limits number of mbuf clusters */
101int nmbjumbop; /* limits number of page size jumbo clusters */
102int nmbjumbo9; /* limits number of 9k jumbo clusters */
103int nmbjumbo16; /* limits number of 16k jumbo clusters */
104struct mbstat mbstat;
105
106/*
107 * tunable_mbinit() has to be run before init_maxsockets() thus
108 * the SYSINIT order below is SI_ORDER_MIDDLE while init_maxsockets()
109 * runs at SI_ORDER_ANY.
110 *
111 * NB: This has to be done before VM init.
112 */
113static void
114tunable_mbinit(void *dummy)
115{
116
117 TUNABLE_INT_FETCH("kern.ipc.nmbclusters", &nmbclusters);
118 if (nmbclusters == 0)
119 nmbclusters = maxmbufmem / MCLBYTES / 4;
120
121 TUNABLE_INT_FETCH("kern.ipc.nmbjumbop", &nmbjumbop);
122 if (nmbjumbop == 0)
123 nmbjumbop = maxmbufmem / MJUMPAGESIZE / 4;
124
125 TUNABLE_INT_FETCH("kern.ipc.nmbjumbo9", &nmbjumbo9);
126 if (nmbjumbo9 == 0)
127 nmbjumbo9 = maxmbufmem / MJUM9BYTES / 6;
128
129 TUNABLE_INT_FETCH("kern.ipc.nmbjumbo16", &nmbjumbo16);
130 if (nmbjumbo16 == 0)
131 nmbjumbo16 = maxmbufmem / MJUM16BYTES / 6;
132
133 /*
134 * We need at least as many mbufs as we have clusters of
135 * the various types added together.
136 */
137 TUNABLE_INT_FETCH("kern.ipc.nmbufs", &nmbufs);
138 if (nmbufs < nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16)
139 nmbufs = lmax(maxmbufmem / MSIZE / 5,
140 nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16);
141
142}
143SYSINIT(tunable_mbinit, SI_SUB_TUNABLES, SI_ORDER_MIDDLE, tunable_mbinit, NULL);
144
145static int
146sysctl_nmbclusters(SYSCTL_HANDLER_ARGS)
147{
148 int error, newnmbclusters;
149
150 newnmbclusters = nmbclusters;
151 error = sysctl_handle_int(oidp, &newnmbclusters, 0, req);
152 if (error == 0 && req->newptr) {
153 if (newnmbclusters > nmbclusters &&
154 nmbufs >= nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16) {
155 nmbclusters = newnmbclusters;
156 uma_zone_set_max(zone_clust, nmbclusters);
157 nmbclusters = uma_zone_get_max(zone_clust);
158 EVENTHANDLER_INVOKE(nmbclusters_change);
159 } else
160 error = EINVAL;
161 }
162 return (error);
163}
164SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbclusters, CTLTYPE_INT|CTLFLAG_RW,
165&nmbclusters, 0, sysctl_nmbclusters, "IU",
166 "Maximum number of mbuf clusters allowed");
167
168static int
169sysctl_nmbjumbop(SYSCTL_HANDLER_ARGS)
170{
171 int error, newnmbjumbop;
172
173 newnmbjumbop = nmbjumbop;
174 error = sysctl_handle_int(oidp, &newnmbjumbop, 0, req);
175 if (error == 0 && req->newptr) {
176 if (newnmbjumbop > nmbjumbop &&
177 nmbufs >= nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16) {
178 nmbjumbop = newnmbjumbop;
179 uma_zone_set_max(zone_jumbop, nmbjumbop);
180 nmbjumbop = uma_zone_get_max(zone_jumbop);
181 } else
182 error = EINVAL;
183 }
184 return (error);
185}
186SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbop, CTLTYPE_INT|CTLFLAG_RW,
187&nmbjumbop, 0, sysctl_nmbjumbop, "IU",
188 "Maximum number of mbuf page size jumbo clusters allowed");
189
190static int
191sysctl_nmbjumbo9(SYSCTL_HANDLER_ARGS)
192{
193 int error, newnmbjumbo9;
194
195 newnmbjumbo9 = nmbjumbo9;
196 error = sysctl_handle_int(oidp, &newnmbjumbo9, 0, req);
197 if (error == 0 && req->newptr) {
198 if (newnmbjumbo9 > nmbjumbo9&&
199 nmbufs >= nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16) {
200 nmbjumbo9 = newnmbjumbo9;
201 uma_zone_set_max(zone_jumbo9, nmbjumbo9);
202 nmbjumbo9 = uma_zone_get_max(zone_jumbo9);
203 } else
204 error = EINVAL;
205 }
206 return (error);
207}
208SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbo9, CTLTYPE_INT|CTLFLAG_RW,
209&nmbjumbo9, 0, sysctl_nmbjumbo9, "IU",
210 "Maximum number of mbuf 9k jumbo clusters allowed");
211
212static int
213sysctl_nmbjumbo16(SYSCTL_HANDLER_ARGS)
214{
215 int error, newnmbjumbo16;
216
217 newnmbjumbo16 = nmbjumbo16;
218 error = sysctl_handle_int(oidp, &newnmbjumbo16, 0, req);
219 if (error == 0 && req->newptr) {
220 if (newnmbjumbo16 > nmbjumbo16 &&
221 nmbufs >= nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16) {
222 nmbjumbo16 = newnmbjumbo16;
223 uma_zone_set_max(zone_jumbo16, nmbjumbo16);
224 nmbjumbo16 = uma_zone_get_max(zone_jumbo16);
225 } else
226 error = EINVAL;
227 }
228 return (error);
229}
230SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbo16, CTLTYPE_INT|CTLFLAG_RW,
231&nmbjumbo16, 0, sysctl_nmbjumbo16, "IU",
232 "Maximum number of mbuf 16k jumbo clusters allowed");
233
234static int
235sysctl_nmbufs(SYSCTL_HANDLER_ARGS)
236{
237 int error, newnmbufs;
238
239 newnmbufs = nmbufs;
240 error = sysctl_handle_int(oidp, &newnmbufs, 0, req);
241 if (error == 0 && req->newptr) {
242 if (newnmbufs > nmbufs) {
243 nmbufs = newnmbufs;
244 uma_zone_set_max(zone_mbuf, nmbufs);
245 nmbufs = uma_zone_get_max(zone_mbuf);
246 EVENTHANDLER_INVOKE(nmbufs_change);
247 } else
248 error = EINVAL;
249 }
250 return (error);
251}
252SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbuf, CTLTYPE_INT|CTLFLAG_RW,
253&nmbufs, 0, sysctl_nmbufs, "IU",
254 "Maximum number of mbufs allowed");
255
256SYSCTL_STRUCT(_kern_ipc, OID_AUTO, mbstat, CTLFLAG_RD, &mbstat, mbstat,
257 "Mbuf general information and statistics");
258
259/*
260 * Zones from which we allocate.
261 */
262uma_zone_t zone_mbuf;
263uma_zone_t zone_clust;
264uma_zone_t zone_pack;
265uma_zone_t zone_jumbop;
266uma_zone_t zone_jumbo9;
267uma_zone_t zone_jumbo16;
268uma_zone_t zone_ext_refcnt;
269
270/*
271 * Local prototypes.
272 */
273static int mb_ctor_mbuf(void *, int, void *, int);
274static int mb_ctor_clust(void *, int, void *, int);
275static int mb_ctor_pack(void *, int, void *, int);
276static void mb_dtor_mbuf(void *, int, void *);
277static void mb_dtor_clust(void *, int, void *);
278static void mb_dtor_pack(void *, int, void *);
279static int mb_zinit_pack(void *, int, int);
280static void mb_zfini_pack(void *, int);
281
282static void mb_reclaim(void *);
283static void mbuf_init(void *);
284static void *mbuf_jumbo_alloc(uma_zone_t, int, uint8_t *, int);
285
286/* Ensure that MSIZE must be a power of 2. */
287CTASSERT((((MSIZE - 1) ^ MSIZE) + 1) >> 1 == MSIZE);
288
289/*
290 * Initialize FreeBSD Network buffer allocation.
291 */
292SYSINIT(mbuf, SI_SUB_MBUF, SI_ORDER_FIRST, mbuf_init, NULL);
293static void
294mbuf_init(void *dummy)
295{
296
297 /*
298 * Configure UMA zones for Mbufs, Clusters, and Packets.
299 */
300 zone_mbuf = uma_zcreate(MBUF_MEM_NAME, MSIZE,
301 mb_ctor_mbuf, mb_dtor_mbuf,
302#ifdef INVARIANTS
303 trash_init, trash_fini,
304#else
305 NULL, NULL,
306#endif
307 MSIZE - 1, UMA_ZONE_MAXBUCKET);
308 if (nmbufs > 0) {
309 uma_zone_set_max(zone_mbuf, nmbufs);
310 nmbufs = uma_zone_get_max(zone_mbuf);
311 }
312
313 zone_clust = uma_zcreate(MBUF_CLUSTER_MEM_NAME, MCLBYTES,
314 mb_ctor_clust, mb_dtor_clust,
315#ifdef INVARIANTS
316 trash_init, trash_fini,
317#else
318 NULL, NULL,
319#endif
320 UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
321 if (nmbclusters > 0) {
322 uma_zone_set_max(zone_clust, nmbclusters);
323 nmbclusters = uma_zone_get_max(zone_clust);
324 }
325
326 zone_pack = uma_zsecond_create(MBUF_PACKET_MEM_NAME, mb_ctor_pack,
327 mb_dtor_pack, mb_zinit_pack, mb_zfini_pack, zone_mbuf);
328
329 /* Make jumbo frame zone too. Page size, 9k and 16k. */
330 zone_jumbop = uma_zcreate(MBUF_JUMBOP_MEM_NAME, MJUMPAGESIZE,
331 mb_ctor_clust, mb_dtor_clust,
332#ifdef INVARIANTS
333 trash_init, trash_fini,
334#else
335 NULL, NULL,
336#endif
337 UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
338 if (nmbjumbop > 0) {
339 uma_zone_set_max(zone_jumbop, nmbjumbop);
340 nmbjumbop = uma_zone_get_max(zone_jumbop);
341 }
342
343 zone_jumbo9 = uma_zcreate(MBUF_JUMBO9_MEM_NAME, MJUM9BYTES,
344 mb_ctor_clust, mb_dtor_clust,
345#ifdef INVARIANTS
346 trash_init, trash_fini,
347#else
348 NULL, NULL,
349#endif
350 UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
351 uma_zone_set_allocf(zone_jumbo9, mbuf_jumbo_alloc);
352 if (nmbjumbo9 > 0) {
353 uma_zone_set_max(zone_jumbo9, nmbjumbo9);
354 nmbjumbo9 = uma_zone_get_max(zone_jumbo9);
355 }
356
357 zone_jumbo16 = uma_zcreate(MBUF_JUMBO16_MEM_NAME, MJUM16BYTES,
358 mb_ctor_clust, mb_dtor_clust,
359#ifdef INVARIANTS
360 trash_init, trash_fini,
361#else
362 NULL, NULL,
363#endif
364 UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
365 uma_zone_set_allocf(zone_jumbo16, mbuf_jumbo_alloc);
366 if (nmbjumbo16 > 0) {
367 uma_zone_set_max(zone_jumbo16, nmbjumbo16);
368 nmbjumbo16 = uma_zone_get_max(zone_jumbo16);
369 }
370
371 zone_ext_refcnt = uma_zcreate(MBUF_EXTREFCNT_MEM_NAME, sizeof(u_int),
372 NULL, NULL,
373 NULL, NULL,
374 UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
375
376 /* uma_prealloc() goes here... */
377
378 /*
379 * Hook event handler for low-memory situation, used to
380 * drain protocols and push data back to the caches (UMA
381 * later pushes it back to VM).
382 */
383 EVENTHANDLER_REGISTER(vm_lowmem, mb_reclaim, NULL,
384 EVENTHANDLER_PRI_FIRST);
385
386 /*
387 * [Re]set counters and local statistics knobs.
388 * XXX Some of these should go and be replaced, but UMA stat
389 * gathering needs to be revised.
390 */
391 mbstat.m_mbufs = 0;
392 mbstat.m_mclusts = 0;
393 mbstat.m_drain = 0;
394 mbstat.m_msize = MSIZE;
395 mbstat.m_mclbytes = MCLBYTES;
396 mbstat.m_minclsize = MINCLSIZE;
397 mbstat.m_mlen = MLEN;
398 mbstat.m_mhlen = MHLEN;
399 mbstat.m_numtypes = MT_NTYPES;
400
401 mbstat.m_mcfail = mbstat.m_mpfail = 0;
402 mbstat.sf_iocnt = 0;
403 mbstat.sf_allocwait = mbstat.sf_allocfail = 0;
404}
405
406/*
407 * UMA backend page allocator for the jumbo frame zones.
408 *
409 * Allocates kernel virtual memory that is backed by contiguous physical
410 * pages.
411 */
412static void *
413mbuf_jumbo_alloc(uma_zone_t zone, int bytes, uint8_t *flags, int wait)
414{
415
416 /* Inform UMA that this allocator uses kernel_map/object. */
417 *flags = UMA_SLAB_KERNEL;
418 return ((void *)kmem_alloc_contig(kernel_map, bytes, wait,
419 (vm_paddr_t)0, ~(vm_paddr_t)0, 1, 0, VM_MEMATTR_DEFAULT));
420}
421
422/*
423 * Constructor for Mbuf master zone.
424 *
425 * The 'arg' pointer points to a mb_args structure which
426 * contains call-specific information required to support the
427 * mbuf allocation API. See mbuf.h.
428 */
429static int
430mb_ctor_mbuf(void *mem, int size, void *arg, int how)
431{
432 struct mbuf *m;
433 struct mb_args *args;
434#ifdef MAC
435 int error;
436#endif
437 int flags;
438 short type;
439
440#ifdef INVARIANTS
441 trash_ctor(mem, size, arg, how);
442#endif
443 m = (struct mbuf *)mem;
444 args = (struct mb_args *)arg;
445 flags = args->flags;
446 type = args->type;
447
448 /*
449 * The mbuf is initialized later. The caller has the
450 * responsibility to set up any MAC labels too.
451 */
452 if (type == MT_NOINIT)
453 return (0);
454
455 m->m_next = NULL;
456 m->m_nextpkt = NULL;
457 m->m_len = 0;
458 m->m_flags = flags;
459 m->m_type = type;
460 if (flags & M_PKTHDR) {
461 m->m_data = m->m_pktdat;
462 m->m_pkthdr.rcvif = NULL;
463 m->m_pkthdr.header = NULL;
464 m->m_pkthdr.len = 0;
465 m->m_pkthdr.csum_flags = 0;
466 m->m_pkthdr.csum_data = 0;
467 m->m_pkthdr.tso_segsz = 0;
468 m->m_pkthdr.ether_vtag = 0;
469 m->m_pkthdr.flowid = 0;
470 SLIST_INIT(&m->m_pkthdr.tags);
471#ifdef MAC
472 /* If the label init fails, fail the alloc */
473 error = mac_mbuf_init(m, how);
474 if (error)
475 return (error);
476#endif
477 } else
478 m->m_data = m->m_dat;
479 return (0);
480}
481
482/*
483 * The Mbuf master zone destructor.
484 */
485static void
486mb_dtor_mbuf(void *mem, int size, void *arg)
487{
488 struct mbuf *m;
489 unsigned long flags;
490
491 m = (struct mbuf *)mem;
492 flags = (unsigned long)arg;
493
494 if ((flags & MB_NOTAGS) == 0 && (m->m_flags & M_PKTHDR) != 0)
495 m_tag_delete_chain(m, NULL);
496 KASSERT((m->m_flags & M_EXT) == 0, ("%s: M_EXT set", __func__));
497 KASSERT((m->m_flags & M_NOFREE) == 0, ("%s: M_NOFREE set", __func__));
498#ifdef INVARIANTS
499 trash_dtor(mem, size, arg);
500#endif
501}
502
503/*
504 * The Mbuf Packet zone destructor.
505 */
506static void
507mb_dtor_pack(void *mem, int size, void *arg)
508{
509 struct mbuf *m;
510
511 m = (struct mbuf *)mem;
512 if ((m->m_flags & M_PKTHDR) != 0)
513 m_tag_delete_chain(m, NULL);
514
515 /* Make sure we've got a clean cluster back. */
516 KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__));
517 KASSERT(m->m_ext.ext_buf != NULL, ("%s: ext_buf == NULL", __func__));
518 KASSERT(m->m_ext.ext_free == NULL, ("%s: ext_free != NULL", __func__));
519 KASSERT(m->m_ext.ext_arg1 == NULL, ("%s: ext_arg1 != NULL", __func__));
520 KASSERT(m->m_ext.ext_arg2 == NULL, ("%s: ext_arg2 != NULL", __func__));
521 KASSERT(m->m_ext.ext_size == MCLBYTES, ("%s: ext_size != MCLBYTES", __func__));
522 KASSERT(m->m_ext.ext_type == EXT_PACKET, ("%s: ext_type != EXT_PACKET", __func__));
523 KASSERT(*m->m_ext.ref_cnt == 1, ("%s: ref_cnt != 1", __func__));
524#ifdef INVARIANTS
525 trash_dtor(m->m_ext.ext_buf, MCLBYTES, arg);
526#endif
527 /*
528 * If there are processes blocked on zone_clust, waiting for pages
529 * to be freed up, * cause them to be woken up by draining the
530 * packet zone. We are exposed to a race here * (in the check for
531 * the UMA_ZFLAG_FULL) where we might miss the flag set, but that
532 * is deliberate. We don't want to acquire the zone lock for every
533 * mbuf free.
534 */
535 if (uma_zone_exhausted_nolock(zone_clust))
536 zone_drain(zone_pack);
537}
538
539/*
540 * The Cluster and Jumbo[PAGESIZE|9|16] zone constructor.
541 *
542 * Here the 'arg' pointer points to the Mbuf which we
543 * are configuring cluster storage for. If 'arg' is
544 * empty we allocate just the cluster without setting
545 * the mbuf to it. See mbuf.h.
546 */
547static int
548mb_ctor_clust(void *mem, int size, void *arg, int how)
549{
550 struct mbuf *m;
551 u_int *refcnt;
552 int type;
553 uma_zone_t zone;
554
555#ifdef INVARIANTS
556 trash_ctor(mem, size, arg, how);
557#endif
558 switch (size) {
559 case MCLBYTES:
560 type = EXT_CLUSTER;
561 zone = zone_clust;
562 break;
563#if MJUMPAGESIZE != MCLBYTES
564 case MJUMPAGESIZE:
565 type = EXT_JUMBOP;
566 zone = zone_jumbop;
567 break;
568#endif
569 case MJUM9BYTES:
570 type = EXT_JUMBO9;
571 zone = zone_jumbo9;
572 break;
573 case MJUM16BYTES:
574 type = EXT_JUMBO16;
575 zone = zone_jumbo16;
576 break;
577 default:
578 panic("unknown cluster size");
579 break;
580 }
581
582 m = (struct mbuf *)arg;
583 refcnt = uma_find_refcnt(zone, mem);
584 *refcnt = 1;
585 if (m != NULL) {
586 m->m_ext.ext_buf = (caddr_t)mem;
587 m->m_data = m->m_ext.ext_buf;
588 m->m_flags |= M_EXT;
589 m->m_ext.ext_free = NULL;
590 m->m_ext.ext_arg1 = NULL;
591 m->m_ext.ext_arg2 = NULL;
592 m->m_ext.ext_size = size;
593 m->m_ext.ext_type = type;
594 m->m_ext.ref_cnt = refcnt;
595 }
596
597 return (0);
598}
599
600/*
601 * The Mbuf Cluster zone destructor.
602 */
603static void
604mb_dtor_clust(void *mem, int size, void *arg)
605{
606#ifdef INVARIANTS
607 uma_zone_t zone;
608
609 zone = m_getzone(size);
610 KASSERT(*(uma_find_refcnt(zone, mem)) <= 1,
611 ("%s: refcnt incorrect %u", __func__,
612 *(uma_find_refcnt(zone, mem))) );
613
614 trash_dtor(mem, size, arg);
615#endif
616}
617
618/*
619 * The Packet secondary zone's init routine, executed on the
620 * object's transition from mbuf keg slab to zone cache.
621 */
622static int
623mb_zinit_pack(void *mem, int size, int how)
624{
625 struct mbuf *m;
626
627 m = (struct mbuf *)mem; /* m is virgin. */
628 if (uma_zalloc_arg(zone_clust, m, how) == NULL ||
629 m->m_ext.ext_buf == NULL)
630 return (ENOMEM);
631 m->m_ext.ext_type = EXT_PACKET; /* Override. */
632#ifdef INVARIANTS
633 trash_init(m->m_ext.ext_buf, MCLBYTES, how);
634#endif
635 return (0);
636}
637
638/*
639 * The Packet secondary zone's fini routine, executed on the
640 * object's transition from zone cache to keg slab.
641 */
642static void
643mb_zfini_pack(void *mem, int size)
644{
645 struct mbuf *m;
646
647 m = (struct mbuf *)mem;
648#ifdef INVARIANTS
649 trash_fini(m->m_ext.ext_buf, MCLBYTES);
650#endif
651 uma_zfree_arg(zone_clust, m->m_ext.ext_buf, NULL);
652#ifdef INVARIANTS
653 trash_dtor(mem, size, NULL);
654#endif
655}
656
657/*
658 * The "packet" keg constructor.
659 */
660static int
661mb_ctor_pack(void *mem, int size, void *arg, int how)
662{
663 struct mbuf *m;
664 struct mb_args *args;
665#ifdef MAC
666 int error;
667#endif
668 int flags;
669 short type;
670
671 m = (struct mbuf *)mem;
672 args = (struct mb_args *)arg;
673 flags = args->flags;
674 type = args->type;
675
676#ifdef INVARIANTS
677 trash_ctor(m->m_ext.ext_buf, MCLBYTES, arg, how);
678#endif
679 m->m_next = NULL;
680 m->m_nextpkt = NULL;
681 m->m_data = m->m_ext.ext_buf;
682 m->m_len = 0;
683 m->m_flags = (flags | M_EXT);
684 m->m_type = type;
685
686 if (flags & M_PKTHDR) {
687 m->m_pkthdr.rcvif = NULL;
688 m->m_pkthdr.len = 0;
689 m->m_pkthdr.header = NULL;
690 m->m_pkthdr.csum_flags = 0;
691 m->m_pkthdr.csum_data = 0;
692 m->m_pkthdr.tso_segsz = 0;
693 m->m_pkthdr.ether_vtag = 0;
694 m->m_pkthdr.flowid = 0;
695 SLIST_INIT(&m->m_pkthdr.tags);
696#ifdef MAC
697 /* If the label init fails, fail the alloc */
698 error = mac_mbuf_init(m, how);
699 if (error)
700 return (error);
701#endif
702 }
703 /* m_ext is already initialized. */
704
705 return (0);
706}
707
708int
709m_pkthdr_init(struct mbuf *m, int how)
710{
711#ifdef MAC
712 int error;
713#endif
714 m->m_data = m->m_pktdat;
715 SLIST_INIT(&m->m_pkthdr.tags);
716 m->m_pkthdr.rcvif = NULL;
717 m->m_pkthdr.header = NULL;
718 m->m_pkthdr.len = 0;
719 m->m_pkthdr.flowid = 0;
720 m->m_pkthdr.csum_flags = 0;
721 m->m_pkthdr.csum_data = 0;
722 m->m_pkthdr.tso_segsz = 0;
723 m->m_pkthdr.ether_vtag = 0;
724#ifdef MAC
725 /* If the label init fails, fail the alloc */
726 error = mac_mbuf_init(m, how);
727 if (error)
728 return (error);
729#endif
730
731 return (0);
732}
733
734/*
735 * This is the protocol drain routine.
736 *
737 * No locks should be held when this is called. The drain routines have to
738 * presently acquire some locks which raises the possibility of lock order
739 * reversal.
740 */
741static void
742mb_reclaim(void *junk)
743{
744 struct domain *dp;
745 struct protosw *pr;
746
747 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK | WARN_PANIC, NULL,
748 "mb_reclaim()");
749
750 for (dp = domains; dp != NULL; dp = dp->dom_next)
751 for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++)
752 if (pr->pr_drain != NULL)
753 (*pr->pr_drain)();
754}
2 * Copyright (c) 2004, 2005,
3 * Bosko Milekic <bmilekic@FreeBSD.org>. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice unmodified, this list of conditions and the following
10 * disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 *
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25 * SUCH DAMAGE.
26 */
27
28#include <sys/cdefs.h>
29__FBSDID("$FreeBSD: head/sys/kern/kern_mbuf.c 243995 2012-12-07 22:19:41Z pjd $");
30
31#include "opt_param.h"
32
33#include <sys/param.h>
34#include <sys/malloc.h>
35#include <sys/systm.h>
36#include <sys/mbuf.h>
37#include <sys/domain.h>
38#include <sys/eventhandler.h>
39#include <sys/kernel.h>
40#include <sys/protosw.h>
41#include <sys/smp.h>
42#include <sys/sysctl.h>
43
44#include <security/mac/mac_framework.h>
45
46#include <vm/vm.h>
47#include <vm/vm_extern.h>
48#include <vm/vm_kern.h>
49#include <vm/vm_page.h>
50#include <vm/uma.h>
51#include <vm/uma_int.h>
52#include <vm/uma_dbg.h>
53
54/*
55 * In FreeBSD, Mbufs and Mbuf Clusters are allocated from UMA
56 * Zones.
57 *
58 * Mbuf Clusters (2K, contiguous) are allocated from the Cluster
59 * Zone. The Zone can be capped at kern.ipc.nmbclusters, if the
60 * administrator so desires.
61 *
62 * Mbufs are allocated from a UMA Master Zone called the Mbuf
63 * Zone.
64 *
65 * Additionally, FreeBSD provides a Packet Zone, which it
66 * configures as a Secondary Zone to the Mbuf Master Zone,
67 * thus sharing backend Slab kegs with the Mbuf Master Zone.
68 *
69 * Thus common-case allocations and locking are simplified:
70 *
71 * m_clget() m_getcl()
72 * | |
73 * | .------------>[(Packet Cache)] m_get(), m_gethdr()
74 * | | [ Packet ] |
75 * [(Cluster Cache)] [ Secondary ] [ (Mbuf Cache) ]
76 * [ Cluster Zone ] [ Zone ] [ Mbuf Master Zone ]
77 * | \________ |
78 * [ Cluster Keg ] \ /
79 * | [ Mbuf Keg ]
80 * [ Cluster Slabs ] |
81 * | [ Mbuf Slabs ]
82 * \____________(VM)_________________/
83 *
84 *
85 * Whenever an object is allocated with uma_zalloc() out of
86 * one of the Zones its _ctor_ function is executed. The same
87 * for any deallocation through uma_zfree() the _dtor_ function
88 * is executed.
89 *
90 * Caches are per-CPU and are filled from the Master Zone.
91 *
92 * Whenever an object is allocated from the underlying global
93 * memory pool it gets pre-initialized with the _zinit_ functions.
94 * When the Keg's are overfull objects get decomissioned with
95 * _zfini_ functions and free'd back to the global memory pool.
96 *
97 */
98
99int nmbufs; /* limits number of mbufs */
100int nmbclusters; /* limits number of mbuf clusters */
101int nmbjumbop; /* limits number of page size jumbo clusters */
102int nmbjumbo9; /* limits number of 9k jumbo clusters */
103int nmbjumbo16; /* limits number of 16k jumbo clusters */
104struct mbstat mbstat;
105
106/*
107 * tunable_mbinit() has to be run before init_maxsockets() thus
108 * the SYSINIT order below is SI_ORDER_MIDDLE while init_maxsockets()
109 * runs at SI_ORDER_ANY.
110 *
111 * NB: This has to be done before VM init.
112 */
113static void
114tunable_mbinit(void *dummy)
115{
116
117 TUNABLE_INT_FETCH("kern.ipc.nmbclusters", &nmbclusters);
118 if (nmbclusters == 0)
119 nmbclusters = maxmbufmem / MCLBYTES / 4;
120
121 TUNABLE_INT_FETCH("kern.ipc.nmbjumbop", &nmbjumbop);
122 if (nmbjumbop == 0)
123 nmbjumbop = maxmbufmem / MJUMPAGESIZE / 4;
124
125 TUNABLE_INT_FETCH("kern.ipc.nmbjumbo9", &nmbjumbo9);
126 if (nmbjumbo9 == 0)
127 nmbjumbo9 = maxmbufmem / MJUM9BYTES / 6;
128
129 TUNABLE_INT_FETCH("kern.ipc.nmbjumbo16", &nmbjumbo16);
130 if (nmbjumbo16 == 0)
131 nmbjumbo16 = maxmbufmem / MJUM16BYTES / 6;
132
133 /*
134 * We need at least as many mbufs as we have clusters of
135 * the various types added together.
136 */
137 TUNABLE_INT_FETCH("kern.ipc.nmbufs", &nmbufs);
138 if (nmbufs < nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16)
139 nmbufs = lmax(maxmbufmem / MSIZE / 5,
140 nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16);
141
142}
143SYSINIT(tunable_mbinit, SI_SUB_TUNABLES, SI_ORDER_MIDDLE, tunable_mbinit, NULL);
144
145static int
146sysctl_nmbclusters(SYSCTL_HANDLER_ARGS)
147{
148 int error, newnmbclusters;
149
150 newnmbclusters = nmbclusters;
151 error = sysctl_handle_int(oidp, &newnmbclusters, 0, req);
152 if (error == 0 && req->newptr) {
153 if (newnmbclusters > nmbclusters &&
154 nmbufs >= nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16) {
155 nmbclusters = newnmbclusters;
156 uma_zone_set_max(zone_clust, nmbclusters);
157 nmbclusters = uma_zone_get_max(zone_clust);
158 EVENTHANDLER_INVOKE(nmbclusters_change);
159 } else
160 error = EINVAL;
161 }
162 return (error);
163}
164SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbclusters, CTLTYPE_INT|CTLFLAG_RW,
165&nmbclusters, 0, sysctl_nmbclusters, "IU",
166 "Maximum number of mbuf clusters allowed");
167
168static int
169sysctl_nmbjumbop(SYSCTL_HANDLER_ARGS)
170{
171 int error, newnmbjumbop;
172
173 newnmbjumbop = nmbjumbop;
174 error = sysctl_handle_int(oidp, &newnmbjumbop, 0, req);
175 if (error == 0 && req->newptr) {
176 if (newnmbjumbop > nmbjumbop &&
177 nmbufs >= nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16) {
178 nmbjumbop = newnmbjumbop;
179 uma_zone_set_max(zone_jumbop, nmbjumbop);
180 nmbjumbop = uma_zone_get_max(zone_jumbop);
181 } else
182 error = EINVAL;
183 }
184 return (error);
185}
186SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbop, CTLTYPE_INT|CTLFLAG_RW,
187&nmbjumbop, 0, sysctl_nmbjumbop, "IU",
188 "Maximum number of mbuf page size jumbo clusters allowed");
189
190static int
191sysctl_nmbjumbo9(SYSCTL_HANDLER_ARGS)
192{
193 int error, newnmbjumbo9;
194
195 newnmbjumbo9 = nmbjumbo9;
196 error = sysctl_handle_int(oidp, &newnmbjumbo9, 0, req);
197 if (error == 0 && req->newptr) {
198 if (newnmbjumbo9 > nmbjumbo9&&
199 nmbufs >= nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16) {
200 nmbjumbo9 = newnmbjumbo9;
201 uma_zone_set_max(zone_jumbo9, nmbjumbo9);
202 nmbjumbo9 = uma_zone_get_max(zone_jumbo9);
203 } else
204 error = EINVAL;
205 }
206 return (error);
207}
208SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbo9, CTLTYPE_INT|CTLFLAG_RW,
209&nmbjumbo9, 0, sysctl_nmbjumbo9, "IU",
210 "Maximum number of mbuf 9k jumbo clusters allowed");
211
212static int
213sysctl_nmbjumbo16(SYSCTL_HANDLER_ARGS)
214{
215 int error, newnmbjumbo16;
216
217 newnmbjumbo16 = nmbjumbo16;
218 error = sysctl_handle_int(oidp, &newnmbjumbo16, 0, req);
219 if (error == 0 && req->newptr) {
220 if (newnmbjumbo16 > nmbjumbo16 &&
221 nmbufs >= nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16) {
222 nmbjumbo16 = newnmbjumbo16;
223 uma_zone_set_max(zone_jumbo16, nmbjumbo16);
224 nmbjumbo16 = uma_zone_get_max(zone_jumbo16);
225 } else
226 error = EINVAL;
227 }
228 return (error);
229}
230SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbo16, CTLTYPE_INT|CTLFLAG_RW,
231&nmbjumbo16, 0, sysctl_nmbjumbo16, "IU",
232 "Maximum number of mbuf 16k jumbo clusters allowed");
233
234static int
235sysctl_nmbufs(SYSCTL_HANDLER_ARGS)
236{
237 int error, newnmbufs;
238
239 newnmbufs = nmbufs;
240 error = sysctl_handle_int(oidp, &newnmbufs, 0, req);
241 if (error == 0 && req->newptr) {
242 if (newnmbufs > nmbufs) {
243 nmbufs = newnmbufs;
244 uma_zone_set_max(zone_mbuf, nmbufs);
245 nmbufs = uma_zone_get_max(zone_mbuf);
246 EVENTHANDLER_INVOKE(nmbufs_change);
247 } else
248 error = EINVAL;
249 }
250 return (error);
251}
252SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbuf, CTLTYPE_INT|CTLFLAG_RW,
253&nmbufs, 0, sysctl_nmbufs, "IU",
254 "Maximum number of mbufs allowed");
255
256SYSCTL_STRUCT(_kern_ipc, OID_AUTO, mbstat, CTLFLAG_RD, &mbstat, mbstat,
257 "Mbuf general information and statistics");
258
259/*
260 * Zones from which we allocate.
261 */
262uma_zone_t zone_mbuf;
263uma_zone_t zone_clust;
264uma_zone_t zone_pack;
265uma_zone_t zone_jumbop;
266uma_zone_t zone_jumbo9;
267uma_zone_t zone_jumbo16;
268uma_zone_t zone_ext_refcnt;
269
270/*
271 * Local prototypes.
272 */
273static int mb_ctor_mbuf(void *, int, void *, int);
274static int mb_ctor_clust(void *, int, void *, int);
275static int mb_ctor_pack(void *, int, void *, int);
276static void mb_dtor_mbuf(void *, int, void *);
277static void mb_dtor_clust(void *, int, void *);
278static void mb_dtor_pack(void *, int, void *);
279static int mb_zinit_pack(void *, int, int);
280static void mb_zfini_pack(void *, int);
281
282static void mb_reclaim(void *);
283static void mbuf_init(void *);
284static void *mbuf_jumbo_alloc(uma_zone_t, int, uint8_t *, int);
285
286/* Ensure that MSIZE must be a power of 2. */
287CTASSERT((((MSIZE - 1) ^ MSIZE) + 1) >> 1 == MSIZE);
288
289/*
290 * Initialize FreeBSD Network buffer allocation.
291 */
292SYSINIT(mbuf, SI_SUB_MBUF, SI_ORDER_FIRST, mbuf_init, NULL);
293static void
294mbuf_init(void *dummy)
295{
296
297 /*
298 * Configure UMA zones for Mbufs, Clusters, and Packets.
299 */
300 zone_mbuf = uma_zcreate(MBUF_MEM_NAME, MSIZE,
301 mb_ctor_mbuf, mb_dtor_mbuf,
302#ifdef INVARIANTS
303 trash_init, trash_fini,
304#else
305 NULL, NULL,
306#endif
307 MSIZE - 1, UMA_ZONE_MAXBUCKET);
308 if (nmbufs > 0) {
309 uma_zone_set_max(zone_mbuf, nmbufs);
310 nmbufs = uma_zone_get_max(zone_mbuf);
311 }
312
313 zone_clust = uma_zcreate(MBUF_CLUSTER_MEM_NAME, MCLBYTES,
314 mb_ctor_clust, mb_dtor_clust,
315#ifdef INVARIANTS
316 trash_init, trash_fini,
317#else
318 NULL, NULL,
319#endif
320 UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
321 if (nmbclusters > 0) {
322 uma_zone_set_max(zone_clust, nmbclusters);
323 nmbclusters = uma_zone_get_max(zone_clust);
324 }
325
326 zone_pack = uma_zsecond_create(MBUF_PACKET_MEM_NAME, mb_ctor_pack,
327 mb_dtor_pack, mb_zinit_pack, mb_zfini_pack, zone_mbuf);
328
329 /* Make jumbo frame zone too. Page size, 9k and 16k. */
330 zone_jumbop = uma_zcreate(MBUF_JUMBOP_MEM_NAME, MJUMPAGESIZE,
331 mb_ctor_clust, mb_dtor_clust,
332#ifdef INVARIANTS
333 trash_init, trash_fini,
334#else
335 NULL, NULL,
336#endif
337 UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
338 if (nmbjumbop > 0) {
339 uma_zone_set_max(zone_jumbop, nmbjumbop);
340 nmbjumbop = uma_zone_get_max(zone_jumbop);
341 }
342
343 zone_jumbo9 = uma_zcreate(MBUF_JUMBO9_MEM_NAME, MJUM9BYTES,
344 mb_ctor_clust, mb_dtor_clust,
345#ifdef INVARIANTS
346 trash_init, trash_fini,
347#else
348 NULL, NULL,
349#endif
350 UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
351 uma_zone_set_allocf(zone_jumbo9, mbuf_jumbo_alloc);
352 if (nmbjumbo9 > 0) {
353 uma_zone_set_max(zone_jumbo9, nmbjumbo9);
354 nmbjumbo9 = uma_zone_get_max(zone_jumbo9);
355 }
356
357 zone_jumbo16 = uma_zcreate(MBUF_JUMBO16_MEM_NAME, MJUM16BYTES,
358 mb_ctor_clust, mb_dtor_clust,
359#ifdef INVARIANTS
360 trash_init, trash_fini,
361#else
362 NULL, NULL,
363#endif
364 UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
365 uma_zone_set_allocf(zone_jumbo16, mbuf_jumbo_alloc);
366 if (nmbjumbo16 > 0) {
367 uma_zone_set_max(zone_jumbo16, nmbjumbo16);
368 nmbjumbo16 = uma_zone_get_max(zone_jumbo16);
369 }
370
371 zone_ext_refcnt = uma_zcreate(MBUF_EXTREFCNT_MEM_NAME, sizeof(u_int),
372 NULL, NULL,
373 NULL, NULL,
374 UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
375
376 /* uma_prealloc() goes here... */
377
378 /*
379 * Hook event handler for low-memory situation, used to
380 * drain protocols and push data back to the caches (UMA
381 * later pushes it back to VM).
382 */
383 EVENTHANDLER_REGISTER(vm_lowmem, mb_reclaim, NULL,
384 EVENTHANDLER_PRI_FIRST);
385
386 /*
387 * [Re]set counters and local statistics knobs.
388 * XXX Some of these should go and be replaced, but UMA stat
389 * gathering needs to be revised.
390 */
391 mbstat.m_mbufs = 0;
392 mbstat.m_mclusts = 0;
393 mbstat.m_drain = 0;
394 mbstat.m_msize = MSIZE;
395 mbstat.m_mclbytes = MCLBYTES;
396 mbstat.m_minclsize = MINCLSIZE;
397 mbstat.m_mlen = MLEN;
398 mbstat.m_mhlen = MHLEN;
399 mbstat.m_numtypes = MT_NTYPES;
400
401 mbstat.m_mcfail = mbstat.m_mpfail = 0;
402 mbstat.sf_iocnt = 0;
403 mbstat.sf_allocwait = mbstat.sf_allocfail = 0;
404}
405
406/*
407 * UMA backend page allocator for the jumbo frame zones.
408 *
409 * Allocates kernel virtual memory that is backed by contiguous physical
410 * pages.
411 */
412static void *
413mbuf_jumbo_alloc(uma_zone_t zone, int bytes, uint8_t *flags, int wait)
414{
415
416 /* Inform UMA that this allocator uses kernel_map/object. */
417 *flags = UMA_SLAB_KERNEL;
418 return ((void *)kmem_alloc_contig(kernel_map, bytes, wait,
419 (vm_paddr_t)0, ~(vm_paddr_t)0, 1, 0, VM_MEMATTR_DEFAULT));
420}
421
422/*
423 * Constructor for Mbuf master zone.
424 *
425 * The 'arg' pointer points to a mb_args structure which
426 * contains call-specific information required to support the
427 * mbuf allocation API. See mbuf.h.
428 */
429static int
430mb_ctor_mbuf(void *mem, int size, void *arg, int how)
431{
432 struct mbuf *m;
433 struct mb_args *args;
434#ifdef MAC
435 int error;
436#endif
437 int flags;
438 short type;
439
440#ifdef INVARIANTS
441 trash_ctor(mem, size, arg, how);
442#endif
443 m = (struct mbuf *)mem;
444 args = (struct mb_args *)arg;
445 flags = args->flags;
446 type = args->type;
447
448 /*
449 * The mbuf is initialized later. The caller has the
450 * responsibility to set up any MAC labels too.
451 */
452 if (type == MT_NOINIT)
453 return (0);
454
455 m->m_next = NULL;
456 m->m_nextpkt = NULL;
457 m->m_len = 0;
458 m->m_flags = flags;
459 m->m_type = type;
460 if (flags & M_PKTHDR) {
461 m->m_data = m->m_pktdat;
462 m->m_pkthdr.rcvif = NULL;
463 m->m_pkthdr.header = NULL;
464 m->m_pkthdr.len = 0;
465 m->m_pkthdr.csum_flags = 0;
466 m->m_pkthdr.csum_data = 0;
467 m->m_pkthdr.tso_segsz = 0;
468 m->m_pkthdr.ether_vtag = 0;
469 m->m_pkthdr.flowid = 0;
470 SLIST_INIT(&m->m_pkthdr.tags);
471#ifdef MAC
472 /* If the label init fails, fail the alloc */
473 error = mac_mbuf_init(m, how);
474 if (error)
475 return (error);
476#endif
477 } else
478 m->m_data = m->m_dat;
479 return (0);
480}
481
482/*
483 * The Mbuf master zone destructor.
484 */
485static void
486mb_dtor_mbuf(void *mem, int size, void *arg)
487{
488 struct mbuf *m;
489 unsigned long flags;
490
491 m = (struct mbuf *)mem;
492 flags = (unsigned long)arg;
493
494 if ((flags & MB_NOTAGS) == 0 && (m->m_flags & M_PKTHDR) != 0)
495 m_tag_delete_chain(m, NULL);
496 KASSERT((m->m_flags & M_EXT) == 0, ("%s: M_EXT set", __func__));
497 KASSERT((m->m_flags & M_NOFREE) == 0, ("%s: M_NOFREE set", __func__));
498#ifdef INVARIANTS
499 trash_dtor(mem, size, arg);
500#endif
501}
502
503/*
504 * The Mbuf Packet zone destructor.
505 */
506static void
507mb_dtor_pack(void *mem, int size, void *arg)
508{
509 struct mbuf *m;
510
511 m = (struct mbuf *)mem;
512 if ((m->m_flags & M_PKTHDR) != 0)
513 m_tag_delete_chain(m, NULL);
514
515 /* Make sure we've got a clean cluster back. */
516 KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__));
517 KASSERT(m->m_ext.ext_buf != NULL, ("%s: ext_buf == NULL", __func__));
518 KASSERT(m->m_ext.ext_free == NULL, ("%s: ext_free != NULL", __func__));
519 KASSERT(m->m_ext.ext_arg1 == NULL, ("%s: ext_arg1 != NULL", __func__));
520 KASSERT(m->m_ext.ext_arg2 == NULL, ("%s: ext_arg2 != NULL", __func__));
521 KASSERT(m->m_ext.ext_size == MCLBYTES, ("%s: ext_size != MCLBYTES", __func__));
522 KASSERT(m->m_ext.ext_type == EXT_PACKET, ("%s: ext_type != EXT_PACKET", __func__));
523 KASSERT(*m->m_ext.ref_cnt == 1, ("%s: ref_cnt != 1", __func__));
524#ifdef INVARIANTS
525 trash_dtor(m->m_ext.ext_buf, MCLBYTES, arg);
526#endif
527 /*
528 * If there are processes blocked on zone_clust, waiting for pages
529 * to be freed up, * cause them to be woken up by draining the
530 * packet zone. We are exposed to a race here * (in the check for
531 * the UMA_ZFLAG_FULL) where we might miss the flag set, but that
532 * is deliberate. We don't want to acquire the zone lock for every
533 * mbuf free.
534 */
535 if (uma_zone_exhausted_nolock(zone_clust))
536 zone_drain(zone_pack);
537}
538
539/*
540 * The Cluster and Jumbo[PAGESIZE|9|16] zone constructor.
541 *
542 * Here the 'arg' pointer points to the Mbuf which we
543 * are configuring cluster storage for. If 'arg' is
544 * empty we allocate just the cluster without setting
545 * the mbuf to it. See mbuf.h.
546 */
547static int
548mb_ctor_clust(void *mem, int size, void *arg, int how)
549{
550 struct mbuf *m;
551 u_int *refcnt;
552 int type;
553 uma_zone_t zone;
554
555#ifdef INVARIANTS
556 trash_ctor(mem, size, arg, how);
557#endif
558 switch (size) {
559 case MCLBYTES:
560 type = EXT_CLUSTER;
561 zone = zone_clust;
562 break;
563#if MJUMPAGESIZE != MCLBYTES
564 case MJUMPAGESIZE:
565 type = EXT_JUMBOP;
566 zone = zone_jumbop;
567 break;
568#endif
569 case MJUM9BYTES:
570 type = EXT_JUMBO9;
571 zone = zone_jumbo9;
572 break;
573 case MJUM16BYTES:
574 type = EXT_JUMBO16;
575 zone = zone_jumbo16;
576 break;
577 default:
578 panic("unknown cluster size");
579 break;
580 }
581
582 m = (struct mbuf *)arg;
583 refcnt = uma_find_refcnt(zone, mem);
584 *refcnt = 1;
585 if (m != NULL) {
586 m->m_ext.ext_buf = (caddr_t)mem;
587 m->m_data = m->m_ext.ext_buf;
588 m->m_flags |= M_EXT;
589 m->m_ext.ext_free = NULL;
590 m->m_ext.ext_arg1 = NULL;
591 m->m_ext.ext_arg2 = NULL;
592 m->m_ext.ext_size = size;
593 m->m_ext.ext_type = type;
594 m->m_ext.ref_cnt = refcnt;
595 }
596
597 return (0);
598}
599
600/*
601 * The Mbuf Cluster zone destructor.
602 */
603static void
604mb_dtor_clust(void *mem, int size, void *arg)
605{
606#ifdef INVARIANTS
607 uma_zone_t zone;
608
609 zone = m_getzone(size);
610 KASSERT(*(uma_find_refcnt(zone, mem)) <= 1,
611 ("%s: refcnt incorrect %u", __func__,
612 *(uma_find_refcnt(zone, mem))) );
613
614 trash_dtor(mem, size, arg);
615#endif
616}
617
618/*
619 * The Packet secondary zone's init routine, executed on the
620 * object's transition from mbuf keg slab to zone cache.
621 */
622static int
623mb_zinit_pack(void *mem, int size, int how)
624{
625 struct mbuf *m;
626
627 m = (struct mbuf *)mem; /* m is virgin. */
628 if (uma_zalloc_arg(zone_clust, m, how) == NULL ||
629 m->m_ext.ext_buf == NULL)
630 return (ENOMEM);
631 m->m_ext.ext_type = EXT_PACKET; /* Override. */
632#ifdef INVARIANTS
633 trash_init(m->m_ext.ext_buf, MCLBYTES, how);
634#endif
635 return (0);
636}
637
638/*
639 * The Packet secondary zone's fini routine, executed on the
640 * object's transition from zone cache to keg slab.
641 */
642static void
643mb_zfini_pack(void *mem, int size)
644{
645 struct mbuf *m;
646
647 m = (struct mbuf *)mem;
648#ifdef INVARIANTS
649 trash_fini(m->m_ext.ext_buf, MCLBYTES);
650#endif
651 uma_zfree_arg(zone_clust, m->m_ext.ext_buf, NULL);
652#ifdef INVARIANTS
653 trash_dtor(mem, size, NULL);
654#endif
655}
656
657/*
658 * The "packet" keg constructor.
659 */
660static int
661mb_ctor_pack(void *mem, int size, void *arg, int how)
662{
663 struct mbuf *m;
664 struct mb_args *args;
665#ifdef MAC
666 int error;
667#endif
668 int flags;
669 short type;
670
671 m = (struct mbuf *)mem;
672 args = (struct mb_args *)arg;
673 flags = args->flags;
674 type = args->type;
675
676#ifdef INVARIANTS
677 trash_ctor(m->m_ext.ext_buf, MCLBYTES, arg, how);
678#endif
679 m->m_next = NULL;
680 m->m_nextpkt = NULL;
681 m->m_data = m->m_ext.ext_buf;
682 m->m_len = 0;
683 m->m_flags = (flags | M_EXT);
684 m->m_type = type;
685
686 if (flags & M_PKTHDR) {
687 m->m_pkthdr.rcvif = NULL;
688 m->m_pkthdr.len = 0;
689 m->m_pkthdr.header = NULL;
690 m->m_pkthdr.csum_flags = 0;
691 m->m_pkthdr.csum_data = 0;
692 m->m_pkthdr.tso_segsz = 0;
693 m->m_pkthdr.ether_vtag = 0;
694 m->m_pkthdr.flowid = 0;
695 SLIST_INIT(&m->m_pkthdr.tags);
696#ifdef MAC
697 /* If the label init fails, fail the alloc */
698 error = mac_mbuf_init(m, how);
699 if (error)
700 return (error);
701#endif
702 }
703 /* m_ext is already initialized. */
704
705 return (0);
706}
707
708int
709m_pkthdr_init(struct mbuf *m, int how)
710{
711#ifdef MAC
712 int error;
713#endif
714 m->m_data = m->m_pktdat;
715 SLIST_INIT(&m->m_pkthdr.tags);
716 m->m_pkthdr.rcvif = NULL;
717 m->m_pkthdr.header = NULL;
718 m->m_pkthdr.len = 0;
719 m->m_pkthdr.flowid = 0;
720 m->m_pkthdr.csum_flags = 0;
721 m->m_pkthdr.csum_data = 0;
722 m->m_pkthdr.tso_segsz = 0;
723 m->m_pkthdr.ether_vtag = 0;
724#ifdef MAC
725 /* If the label init fails, fail the alloc */
726 error = mac_mbuf_init(m, how);
727 if (error)
728 return (error);
729#endif
730
731 return (0);
732}
733
734/*
735 * This is the protocol drain routine.
736 *
737 * No locks should be held when this is called. The drain routines have to
738 * presently acquire some locks which raises the possibility of lock order
739 * reversal.
740 */
741static void
742mb_reclaim(void *junk)
743{
744 struct domain *dp;
745 struct protosw *pr;
746
747 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK | WARN_PANIC, NULL,
748 "mb_reclaim()");
749
750 for (dp = domains; dp != NULL; dp = dp->dom_next)
751 for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++)
752 if (pr->pr_drain != NULL)
753 (*pr->pr_drain)();
754}