xref: /freebsd/sys/kern/kern_mbuf.c (revision 2be1a816b9ff69588e55be0a84cbe2a31efc0f2f)
1 /*-
2  * Copyright (c) 2004, 2005,
3  * 	Bosko Milekic <bmilekic@FreeBSD.org>.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice unmodified, this list of conditions and the following
10  *    disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 #include "opt_mac.h"
32 #include "opt_param.h"
33 
34 #include <sys/param.h>
35 #include <sys/malloc.h>
36 #include <sys/systm.h>
37 #include <sys/mbuf.h>
38 #include <sys/domain.h>
39 #include <sys/eventhandler.h>
40 #include <sys/kernel.h>
41 #include <sys/protosw.h>
42 #include <sys/smp.h>
43 #include <sys/sysctl.h>
44 
45 #include <security/mac/mac_framework.h>
46 
47 #include <vm/vm.h>
48 #include <vm/vm_page.h>
49 #include <vm/uma.h>
50 #include <vm/uma_int.h>
51 #include <vm/uma_dbg.h>
52 
53 /*
54  * In FreeBSD, Mbufs and Mbuf Clusters are allocated from UMA
55  * Zones.
56  *
57  * Mbuf Clusters (2K, contiguous) are allocated from the Cluster
58  * Zone.  The Zone can be capped at kern.ipc.nmbclusters, if the
59  * administrator so desires.
60  *
61  * Mbufs are allocated from a UMA Master Zone called the Mbuf
62  * Zone.
63  *
64  * Additionally, FreeBSD provides a Packet Zone, which it
65  * configures as a Secondary Zone to the Mbuf Master Zone,
66  * thus sharing backend Slab kegs with the Mbuf Master Zone.
67  *
68  * Thus common-case allocations and locking are simplified:
69  *
70  *  m_clget()                m_getcl()
71  *    |                         |
72  *    |   .------------>[(Packet Cache)]    m_get(), m_gethdr()
73  *    |   |             [     Packet   ]            |
74  *  [(Cluster Cache)]   [    Secondary ]   [ (Mbuf Cache)     ]
75  *  [ Cluster Zone  ]   [     Zone     ]   [ Mbuf Master Zone ]
76  *        |                       \________         |
77  *  [ Cluster Keg   ]                      \       /
78  *        |    	                         [ Mbuf Keg   ]
79  *  [ Cluster Slabs ]                         |
80  *        |                              [ Mbuf Slabs ]
81  *         \____________(VM)_________________/
82  *
83  *
84  * Whenever an object is allocated with uma_zalloc() out of
85  * one of the Zones its _ctor_ function is executed.  The same
86  * for any deallocation through uma_zfree() the _dtor_ function
87  * is executed.
88  *
89  * Caches are per-CPU and are filled from the Master Zone.
90  *
91  * Whenever an object is allocated from the underlying global
92  * memory pool it gets pre-initialized with the _zinit_ functions.
93  * When the Keg's are overfull objects get decomissioned with
94  * _zfini_ functions and free'd back to the global memory pool.
95  *
96  */
97 
98 int nmbclusters;		/* limits number of mbuf clusters */
99 int nmbjumbop;			/* limits number of page size jumbo clusters */
100 int nmbjumbo9;			/* limits number of 9k jumbo clusters */
101 int nmbjumbo16;			/* limits number of 16k jumbo clusters */
102 struct mbstat mbstat;
103 
104 static void
105 tunable_mbinit(void *dummy)
106 {
107 
108 	/* This has to be done before VM init. */
109 	nmbclusters = 1024 + maxusers * 64;
110 	nmbjumbop = nmbclusters / 2;
111 	nmbjumbo9 = nmbjumbop / 2;
112 	nmbjumbo16 = nmbjumbo9 / 2;
113 	TUNABLE_INT_FETCH("kern.ipc.nmbclusters", &nmbclusters);
114 }
115 SYSINIT(tunable_mbinit, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_mbinit, NULL);
116 
117 /* XXX: These should be tuneables. Can't change UMA limits on the fly. */
118 static int
119 sysctl_nmbclusters(SYSCTL_HANDLER_ARGS)
120 {
121 	int error, newnmbclusters;
122 
123 	newnmbclusters = nmbclusters;
124 	error = sysctl_handle_int(oidp, &newnmbclusters, 0, req);
125 	if (error == 0 && req->newptr) {
126 		if (newnmbclusters > nmbclusters) {
127 			nmbclusters = newnmbclusters;
128 			uma_zone_set_max(zone_clust, nmbclusters);
129 			EVENTHANDLER_INVOKE(nmbclusters_change);
130 		} else
131 			error = EINVAL;
132 	}
133 	return (error);
134 }
135 SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbclusters, CTLTYPE_INT|CTLFLAG_RW,
136 &nmbclusters, 0, sysctl_nmbclusters, "IU",
137     "Maximum number of mbuf clusters allowed");
138 
139 static int
140 sysctl_nmbjumbop(SYSCTL_HANDLER_ARGS)
141 {
142 	int error, newnmbjumbop;
143 
144 	newnmbjumbop = nmbjumbop;
145 	error = sysctl_handle_int(oidp, &newnmbjumbop, 0, req);
146 	if (error == 0 && req->newptr) {
147 		if (newnmbjumbop> nmbjumbop) {
148 			nmbjumbop = newnmbjumbop;
149 			uma_zone_set_max(zone_jumbop, nmbjumbop);
150 		} else
151 			error = EINVAL;
152 	}
153 	return (error);
154 }
155 SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbop, CTLTYPE_INT|CTLFLAG_RW,
156 &nmbjumbop, 0, sysctl_nmbjumbop, "IU",
157 	 "Maximum number of mbuf page size jumbo clusters allowed");
158 
159 
160 static int
161 sysctl_nmbjumbo9(SYSCTL_HANDLER_ARGS)
162 {
163 	int error, newnmbjumbo9;
164 
165 	newnmbjumbo9 = nmbjumbo9;
166 	error = sysctl_handle_int(oidp, &newnmbjumbo9, 0, req);
167 	if (error == 0 && req->newptr) {
168 		if (newnmbjumbo9> nmbjumbo9) {
169 			nmbjumbo9 = newnmbjumbo9;
170 			uma_zone_set_max(zone_jumbo9, nmbjumbo9);
171 		} else
172 			error = EINVAL;
173 	}
174 	return (error);
175 }
176 SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbo9, CTLTYPE_INT|CTLFLAG_RW,
177 &nmbjumbo9, 0, sysctl_nmbjumbo9, "IU",
178 	"Maximum number of mbuf 9k jumbo clusters allowed");
179 
180 static int
181 sysctl_nmbjumbo16(SYSCTL_HANDLER_ARGS)
182 {
183 	int error, newnmbjumbo16;
184 
185 	newnmbjumbo16 = nmbjumbo16;
186 	error = sysctl_handle_int(oidp, &newnmbjumbo16, 0, req);
187 	if (error == 0 && req->newptr) {
188 		if (newnmbjumbo16> nmbjumbo16) {
189 			nmbjumbo16 = newnmbjumbo16;
190 			uma_zone_set_max(zone_jumbo16, nmbjumbo16);
191 		} else
192 			error = EINVAL;
193 	}
194 	return (error);
195 }
196 SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbo16, CTLTYPE_INT|CTLFLAG_RW,
197 &nmbjumbo16, 0, sysctl_nmbjumbo16, "IU",
198     "Maximum number of mbuf 16k jumbo clusters allowed");
199 
200 
201 
202 SYSCTL_STRUCT(_kern_ipc, OID_AUTO, mbstat, CTLFLAG_RD, &mbstat, mbstat,
203     "Mbuf general information and statistics");
204 
205 /*
206  * Zones from which we allocate.
207  */
208 uma_zone_t	zone_mbuf;
209 uma_zone_t	zone_clust;
210 uma_zone_t	zone_pack;
211 uma_zone_t	zone_jumbop;
212 uma_zone_t	zone_jumbo9;
213 uma_zone_t	zone_jumbo16;
214 uma_zone_t	zone_ext_refcnt;
215 
216 /*
217  * Local prototypes.
218  */
219 static int	mb_ctor_mbuf(void *, int, void *, int);
220 static int	mb_ctor_clust(void *, int, void *, int);
221 static int	mb_ctor_pack(void *, int, void *, int);
222 static void	mb_dtor_mbuf(void *, int, void *);
223 static void	mb_dtor_clust(void *, int, void *);
224 static void	mb_dtor_pack(void *, int, void *);
225 static int	mb_zinit_pack(void *, int, int);
226 static void	mb_zfini_pack(void *, int);
227 
228 static void	mb_reclaim(void *);
229 static void	mbuf_init(void *);
230 static void    *mbuf_jumbo_alloc(uma_zone_t, int, u_int8_t *, int);
231 static void	mbuf_jumbo_free(void *, int, u_int8_t);
232 
233 static MALLOC_DEFINE(M_JUMBOFRAME, "jumboframes", "mbuf jumbo frame buffers");
234 
235 /* Ensure that MSIZE doesn't break dtom() - it must be a power of 2 */
236 CTASSERT((((MSIZE - 1) ^ MSIZE) + 1) >> 1 == MSIZE);
237 
238 /*
239  * Initialize FreeBSD Network buffer allocation.
240  */
241 SYSINIT(mbuf, SI_SUB_MBUF, SI_ORDER_FIRST, mbuf_init, NULL);
242 static void
243 mbuf_init(void *dummy)
244 {
245 
246 	/*
247 	 * Configure UMA zones for Mbufs, Clusters, and Packets.
248 	 */
249 	zone_mbuf = uma_zcreate(MBUF_MEM_NAME, MSIZE,
250 	    mb_ctor_mbuf, mb_dtor_mbuf,
251 #ifdef INVARIANTS
252 	    trash_init, trash_fini,
253 #else
254 	    NULL, NULL,
255 #endif
256 	    MSIZE - 1, UMA_ZONE_MAXBUCKET);
257 
258 	zone_clust = uma_zcreate(MBUF_CLUSTER_MEM_NAME, MCLBYTES,
259 	    mb_ctor_clust, mb_dtor_clust,
260 #ifdef INVARIANTS
261 	    trash_init, trash_fini,
262 #else
263 	    NULL, NULL,
264 #endif
265 	    UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
266 	if (nmbclusters > 0)
267 		uma_zone_set_max(zone_clust, nmbclusters);
268 
269 	zone_pack = uma_zsecond_create(MBUF_PACKET_MEM_NAME, mb_ctor_pack,
270 	    mb_dtor_pack, mb_zinit_pack, mb_zfini_pack, zone_mbuf);
271 
272 	/* Make jumbo frame zone too. Page size, 9k and 16k. */
273 	zone_jumbop = uma_zcreate(MBUF_JUMBOP_MEM_NAME, MJUMPAGESIZE,
274 	    mb_ctor_clust, mb_dtor_clust,
275 #ifdef INVARIANTS
276 	    trash_init, trash_fini,
277 #else
278 	    NULL, NULL,
279 #endif
280 	    UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
281 	if (nmbjumbop > 0)
282 		uma_zone_set_max(zone_jumbop, nmbjumbop);
283 
284 	zone_jumbo9 = uma_zcreate(MBUF_JUMBO9_MEM_NAME, MJUM9BYTES,
285 	    mb_ctor_clust, mb_dtor_clust,
286 #ifdef INVARIANTS
287 	    trash_init, trash_fini,
288 #else
289 	    NULL, NULL,
290 #endif
291 	    UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
292 	if (nmbjumbo9 > 0)
293 		uma_zone_set_max(zone_jumbo9, nmbjumbo9);
294 	uma_zone_set_allocf(zone_jumbo9, mbuf_jumbo_alloc);
295 	uma_zone_set_freef(zone_jumbo9, mbuf_jumbo_free);
296 
297 	zone_jumbo16 = uma_zcreate(MBUF_JUMBO16_MEM_NAME, MJUM16BYTES,
298 	    mb_ctor_clust, mb_dtor_clust,
299 #ifdef INVARIANTS
300 	    trash_init, trash_fini,
301 #else
302 	    NULL, NULL,
303 #endif
304 	    UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
305 	if (nmbjumbo16 > 0)
306 		uma_zone_set_max(zone_jumbo16, nmbjumbo16);
307 	uma_zone_set_allocf(zone_jumbo16, mbuf_jumbo_alloc);
308 	uma_zone_set_freef(zone_jumbo16, mbuf_jumbo_free);
309 
310 	zone_ext_refcnt = uma_zcreate(MBUF_EXTREFCNT_MEM_NAME, sizeof(u_int),
311 	    NULL, NULL,
312 	    NULL, NULL,
313 	    UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
314 
315 	/* uma_prealloc() goes here... */
316 
317 	/*
318 	 * Hook event handler for low-memory situation, used to
319 	 * drain protocols and push data back to the caches (UMA
320 	 * later pushes it back to VM).
321 	 */
322 	EVENTHANDLER_REGISTER(vm_lowmem, mb_reclaim, NULL,
323 	    EVENTHANDLER_PRI_FIRST);
324 
325 	/*
326 	 * [Re]set counters and local statistics knobs.
327 	 * XXX Some of these should go and be replaced, but UMA stat
328 	 * gathering needs to be revised.
329 	 */
330 	mbstat.m_mbufs = 0;
331 	mbstat.m_mclusts = 0;
332 	mbstat.m_drain = 0;
333 	mbstat.m_msize = MSIZE;
334 	mbstat.m_mclbytes = MCLBYTES;
335 	mbstat.m_minclsize = MINCLSIZE;
336 	mbstat.m_mlen = MLEN;
337 	mbstat.m_mhlen = MHLEN;
338 	mbstat.m_numtypes = MT_NTYPES;
339 
340 	mbstat.m_mcfail = mbstat.m_mpfail = 0;
341 	mbstat.sf_iocnt = 0;
342 	mbstat.sf_allocwait = mbstat.sf_allocfail = 0;
343 }
344 
345 /*
346  * UMA backend page allocator for the jumbo frame zones.
347  *
348  * Allocates kernel virtual memory that is backed by contiguous physical
349  * pages.
350  */
351 static void *
352 mbuf_jumbo_alloc(uma_zone_t zone, int bytes, u_int8_t *flags, int wait)
353 {
354 
355 	/* Inform UMA that this allocator uses kernel_map/object. */
356 	*flags = UMA_SLAB_KERNEL;
357 	return (contigmalloc(bytes, M_JUMBOFRAME, wait, (vm_paddr_t)0,
358 	    ~(vm_paddr_t)0, 1, 0));
359 }
360 
361 /*
362  * UMA backend page deallocator for the jumbo frame zones.
363  */
364 static void
365 mbuf_jumbo_free(void *mem, int size, u_int8_t flags)
366 {
367 
368 	contigfree(mem, size, M_JUMBOFRAME);
369 }
370 
371 /*
372  * Constructor for Mbuf master zone.
373  *
374  * The 'arg' pointer points to a mb_args structure which
375  * contains call-specific information required to support the
376  * mbuf allocation API.  See mbuf.h.
377  */
378 static int
379 mb_ctor_mbuf(void *mem, int size, void *arg, int how)
380 {
381 	struct mbuf *m;
382 	struct mb_args *args;
383 #ifdef MAC
384 	int error;
385 #endif
386 	int flags;
387 	short type;
388 
389 #ifdef INVARIANTS
390 	trash_ctor(mem, size, arg, how);
391 #endif
392 	m = (struct mbuf *)mem;
393 	args = (struct mb_args *)arg;
394 	flags = args->flags;
395 	type = args->type;
396 
397 	/*
398 	 * The mbuf is initialized later.  The caller has the
399 	 * responsibility to set up any MAC labels too.
400 	 */
401 	if (type == MT_NOINIT)
402 		return (0);
403 
404 	m->m_next = NULL;
405 	m->m_nextpkt = NULL;
406 	m->m_len = 0;
407 	m->m_flags = flags;
408 	m->m_type = type;
409 	if (flags & M_PKTHDR) {
410 		m->m_data = m->m_pktdat;
411 		m->m_pkthdr.rcvif = NULL;
412 		m->m_pkthdr.len = 0;
413 		m->m_pkthdr.header = NULL;
414 		m->m_pkthdr.csum_flags = 0;
415 		m->m_pkthdr.csum_data = 0;
416 		m->m_pkthdr.tso_segsz = 0;
417 		m->m_pkthdr.ether_vtag = 0;
418 		SLIST_INIT(&m->m_pkthdr.tags);
419 #ifdef MAC
420 		/* If the label init fails, fail the alloc */
421 		error = mac_mbuf_init(m, how);
422 		if (error)
423 			return (error);
424 #endif
425 	} else
426 		m->m_data = m->m_dat;
427 	return (0);
428 }
429 
430 /*
431  * The Mbuf master zone destructor.
432  */
433 static void
434 mb_dtor_mbuf(void *mem, int size, void *arg)
435 {
436 	struct mbuf *m;
437 	unsigned long flags;
438 
439 	m = (struct mbuf *)mem;
440 	flags = (unsigned long)arg;
441 
442 	if ((flags & MB_NOTAGS) == 0 && (m->m_flags & M_PKTHDR) != 0)
443 		m_tag_delete_chain(m, NULL);
444 	KASSERT((m->m_flags & M_EXT) == 0, ("%s: M_EXT set", __func__));
445 	KASSERT((m->m_flags & M_NOFREE) == 0, ("%s: M_NOFREE set", __func__));
446 #ifdef INVARIANTS
447 	trash_dtor(mem, size, arg);
448 #endif
449 }
450 
451 /*
452  * The Mbuf Packet zone destructor.
453  */
454 static void
455 mb_dtor_pack(void *mem, int size, void *arg)
456 {
457 	struct mbuf *m;
458 
459 	m = (struct mbuf *)mem;
460 	if ((m->m_flags & M_PKTHDR) != 0)
461 		m_tag_delete_chain(m, NULL);
462 
463 	/* Make sure we've got a clean cluster back. */
464 	KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__));
465 	KASSERT(m->m_ext.ext_buf != NULL, ("%s: ext_buf == NULL", __func__));
466 	KASSERT(m->m_ext.ext_free == NULL, ("%s: ext_free != NULL", __func__));
467 	KASSERT(m->m_ext.ext_arg1 == NULL, ("%s: ext_arg1 != NULL", __func__));
468 	KASSERT(m->m_ext.ext_arg2 == NULL, ("%s: ext_arg2 != NULL", __func__));
469 	KASSERT(m->m_ext.ext_size == MCLBYTES, ("%s: ext_size != MCLBYTES", __func__));
470 	KASSERT(m->m_ext.ext_type == EXT_PACKET, ("%s: ext_type != EXT_PACKET", __func__));
471 	KASSERT(*m->m_ext.ref_cnt == 1, ("%s: ref_cnt != 1", __func__));
472 #ifdef INVARIANTS
473 	trash_dtor(m->m_ext.ext_buf, MCLBYTES, arg);
474 #endif
475 	/*
476 	 * If there are processes blocked on zone_clust, waiting for pages
477 	 * to be freed up, * cause them to be woken up by draining the
478 	 * packet zone.  We are exposed to a race here * (in the check for
479 	 * the UMA_ZFLAG_FULL) where we might miss the flag set, but that
480 	 * is deliberate. We don't want to acquire the zone lock for every
481 	 * mbuf free.
482 	 */
483 	if (uma_zone_exhausted_nolock(zone_clust))
484 		zone_drain(zone_pack);
485 }
486 
487 /*
488  * The Cluster and Jumbo[PAGESIZE|9|16] zone constructor.
489  *
490  * Here the 'arg' pointer points to the Mbuf which we
491  * are configuring cluster storage for.  If 'arg' is
492  * empty we allocate just the cluster without setting
493  * the mbuf to it.  See mbuf.h.
494  */
495 static int
496 mb_ctor_clust(void *mem, int size, void *arg, int how)
497 {
498 	struct mbuf *m;
499 	u_int *refcnt;
500 	int type;
501 	uma_zone_t zone;
502 
503 #ifdef INVARIANTS
504 	trash_ctor(mem, size, arg, how);
505 #endif
506 	switch (size) {
507 	case MCLBYTES:
508 		type = EXT_CLUSTER;
509 		zone = zone_clust;
510 		break;
511 #if MJUMPAGESIZE != MCLBYTES
512 	case MJUMPAGESIZE:
513 		type = EXT_JUMBOP;
514 		zone = zone_jumbop;
515 		break;
516 #endif
517 	case MJUM9BYTES:
518 		type = EXT_JUMBO9;
519 		zone = zone_jumbo9;
520 		break;
521 	case MJUM16BYTES:
522 		type = EXT_JUMBO16;
523 		zone = zone_jumbo16;
524 		break;
525 	default:
526 		panic("unknown cluster size");
527 		break;
528 	}
529 
530 	m = (struct mbuf *)arg;
531 	refcnt = uma_find_refcnt(zone, mem);
532 	*refcnt = 1;
533 	if (m != NULL) {
534 		m->m_ext.ext_buf = (caddr_t)mem;
535 		m->m_data = m->m_ext.ext_buf;
536 		m->m_flags |= M_EXT;
537 		m->m_ext.ext_free = NULL;
538 		m->m_ext.ext_arg1 = NULL;
539 		m->m_ext.ext_arg2 = NULL;
540 		m->m_ext.ext_size = size;
541 		m->m_ext.ext_type = type;
542 		m->m_ext.ref_cnt = refcnt;
543 	}
544 
545 	return (0);
546 }
547 
548 /*
549  * The Mbuf Cluster zone destructor.
550  */
551 static void
552 mb_dtor_clust(void *mem, int size, void *arg)
553 {
554 #ifdef INVARIANTS
555 	uma_zone_t zone;
556 
557 	zone = m_getzone(size);
558 	KASSERT(*(uma_find_refcnt(zone, mem)) <= 1,
559 		("%s: refcnt incorrect %u", __func__,
560 		 *(uma_find_refcnt(zone, mem))) );
561 
562 	trash_dtor(mem, size, arg);
563 #endif
564 }
565 
566 /*
567  * The Packet secondary zone's init routine, executed on the
568  * object's transition from mbuf keg slab to zone cache.
569  */
570 static int
571 mb_zinit_pack(void *mem, int size, int how)
572 {
573 	struct mbuf *m;
574 
575 	m = (struct mbuf *)mem;		/* m is virgin. */
576 	if (uma_zalloc_arg(zone_clust, m, how) == NULL ||
577 	    m->m_ext.ext_buf == NULL)
578 		return (ENOMEM);
579 	m->m_ext.ext_type = EXT_PACKET;	/* Override. */
580 #ifdef INVARIANTS
581 	trash_init(m->m_ext.ext_buf, MCLBYTES, how);
582 #endif
583 	return (0);
584 }
585 
586 /*
587  * The Packet secondary zone's fini routine, executed on the
588  * object's transition from zone cache to keg slab.
589  */
590 static void
591 mb_zfini_pack(void *mem, int size)
592 {
593 	struct mbuf *m;
594 
595 	m = (struct mbuf *)mem;
596 #ifdef INVARIANTS
597 	trash_fini(m->m_ext.ext_buf, MCLBYTES);
598 #endif
599 	uma_zfree_arg(zone_clust, m->m_ext.ext_buf, NULL);
600 #ifdef INVARIANTS
601 	trash_dtor(mem, size, NULL);
602 #endif
603 }
604 
605 /*
606  * The "packet" keg constructor.
607  */
608 static int
609 mb_ctor_pack(void *mem, int size, void *arg, int how)
610 {
611 	struct mbuf *m;
612 	struct mb_args *args;
613 #ifdef MAC
614 	int error;
615 #endif
616 	int flags;
617 	short type;
618 
619 	m = (struct mbuf *)mem;
620 	args = (struct mb_args *)arg;
621 	flags = args->flags;
622 	type = args->type;
623 
624 #ifdef INVARIANTS
625 	trash_ctor(m->m_ext.ext_buf, MCLBYTES, arg, how);
626 #endif
627 	m->m_next = NULL;
628 	m->m_nextpkt = NULL;
629 	m->m_data = m->m_ext.ext_buf;
630 	m->m_len = 0;
631 	m->m_flags = (flags | M_EXT);
632 	m->m_type = type;
633 
634 	if (flags & M_PKTHDR) {
635 		m->m_pkthdr.rcvif = NULL;
636 		m->m_pkthdr.len = 0;
637 		m->m_pkthdr.header = NULL;
638 		m->m_pkthdr.csum_flags = 0;
639 		m->m_pkthdr.csum_data = 0;
640 		m->m_pkthdr.tso_segsz = 0;
641 		m->m_pkthdr.ether_vtag = 0;
642 		SLIST_INIT(&m->m_pkthdr.tags);
643 #ifdef MAC
644 		/* If the label init fails, fail the alloc */
645 		error = mac_mbuf_init(m, how);
646 		if (error)
647 			return (error);
648 #endif
649 	}
650 	/* m_ext is already initialized. */
651 
652 	return (0);
653 }
654 
655 /*
656  * This is the protocol drain routine.
657  *
658  * No locks should be held when this is called.  The drain routines have to
659  * presently acquire some locks which raises the possibility of lock order
660  * reversal.
661  */
662 static void
663 mb_reclaim(void *junk)
664 {
665 	struct domain *dp;
666 	struct protosw *pr;
667 
668 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK | WARN_PANIC, NULL,
669 	    "mb_reclaim()");
670 
671 	for (dp = domains; dp != NULL; dp = dp->dom_next)
672 		for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++)
673 			if (pr->pr_drain != NULL)
674 				(*pr->pr_drain)();
675 }
676