xref: /freebsd/sys/kern/kern_mbuf.c (revision 54ebdd631db8c0bba2baab0155f603a8b5cf014a)
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 /*
105  * tunable_mbinit() has to be run before init_maxsockets() thus
106  * the SYSINIT order below is SI_ORDER_MIDDLE while init_maxsockets()
107  * runs at SI_ORDER_ANY.
108  */
109 static void
110 tunable_mbinit(void *dummy)
111 {
112 	TUNABLE_INT_FETCH("kern.ipc.nmbclusters", &nmbclusters);
113 
114 	/* This has to be done before VM init. */
115 	if (nmbclusters == 0)
116 		nmbclusters = 1024 + maxusers * 64;
117 	nmbjumbop = nmbclusters / 2;
118 	nmbjumbo9 = nmbjumbop / 2;
119 	nmbjumbo16 = nmbjumbo9 / 2;
120 }
121 SYSINIT(tunable_mbinit, SI_SUB_TUNABLES, SI_ORDER_MIDDLE, tunable_mbinit, NULL);
122 
123 /* XXX: These should be tuneables. Can't change UMA limits on the fly. */
124 static int
125 sysctl_nmbclusters(SYSCTL_HANDLER_ARGS)
126 {
127 	int error, newnmbclusters;
128 
129 	newnmbclusters = nmbclusters;
130 	error = sysctl_handle_int(oidp, &newnmbclusters, 0, req);
131 	if (error == 0 && req->newptr) {
132 		if (newnmbclusters > nmbclusters) {
133 			nmbclusters = newnmbclusters;
134 			uma_zone_set_max(zone_clust, nmbclusters);
135 			EVENTHANDLER_INVOKE(nmbclusters_change);
136 		} else
137 			error = EINVAL;
138 	}
139 	return (error);
140 }
141 SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbclusters, CTLTYPE_INT|CTLFLAG_RW,
142 &nmbclusters, 0, sysctl_nmbclusters, "IU",
143     "Maximum number of mbuf clusters allowed");
144 
145 static int
146 sysctl_nmbjumbop(SYSCTL_HANDLER_ARGS)
147 {
148 	int error, newnmbjumbop;
149 
150 	newnmbjumbop = nmbjumbop;
151 	error = sysctl_handle_int(oidp, &newnmbjumbop, 0, req);
152 	if (error == 0 && req->newptr) {
153 		if (newnmbjumbop> nmbjumbop) {
154 			nmbjumbop = newnmbjumbop;
155 			uma_zone_set_max(zone_jumbop, nmbjumbop);
156 		} else
157 			error = EINVAL;
158 	}
159 	return (error);
160 }
161 SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbop, CTLTYPE_INT|CTLFLAG_RW,
162 &nmbjumbop, 0, sysctl_nmbjumbop, "IU",
163 	 "Maximum number of mbuf page size jumbo clusters allowed");
164 
165 
166 static int
167 sysctl_nmbjumbo9(SYSCTL_HANDLER_ARGS)
168 {
169 	int error, newnmbjumbo9;
170 
171 	newnmbjumbo9 = nmbjumbo9;
172 	error = sysctl_handle_int(oidp, &newnmbjumbo9, 0, req);
173 	if (error == 0 && req->newptr) {
174 		if (newnmbjumbo9> nmbjumbo9) {
175 			nmbjumbo9 = newnmbjumbo9;
176 			uma_zone_set_max(zone_jumbo9, nmbjumbo9);
177 		} else
178 			error = EINVAL;
179 	}
180 	return (error);
181 }
182 SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbo9, CTLTYPE_INT|CTLFLAG_RW,
183 &nmbjumbo9, 0, sysctl_nmbjumbo9, "IU",
184 	"Maximum number of mbuf 9k jumbo clusters allowed");
185 
186 static int
187 sysctl_nmbjumbo16(SYSCTL_HANDLER_ARGS)
188 {
189 	int error, newnmbjumbo16;
190 
191 	newnmbjumbo16 = nmbjumbo16;
192 	error = sysctl_handle_int(oidp, &newnmbjumbo16, 0, req);
193 	if (error == 0 && req->newptr) {
194 		if (newnmbjumbo16> nmbjumbo16) {
195 			nmbjumbo16 = newnmbjumbo16;
196 			uma_zone_set_max(zone_jumbo16, nmbjumbo16);
197 		} else
198 			error = EINVAL;
199 	}
200 	return (error);
201 }
202 SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbo16, CTLTYPE_INT|CTLFLAG_RW,
203 &nmbjumbo16, 0, sysctl_nmbjumbo16, "IU",
204     "Maximum number of mbuf 16k jumbo clusters allowed");
205 
206 
207 
208 SYSCTL_STRUCT(_kern_ipc, OID_AUTO, mbstat, CTLFLAG_RD, &mbstat, mbstat,
209     "Mbuf general information and statistics");
210 
211 /*
212  * Zones from which we allocate.
213  */
214 uma_zone_t	zone_mbuf;
215 uma_zone_t	zone_clust;
216 uma_zone_t	zone_pack;
217 uma_zone_t	zone_jumbop;
218 uma_zone_t	zone_jumbo9;
219 uma_zone_t	zone_jumbo16;
220 uma_zone_t	zone_ext_refcnt;
221 
222 /*
223  * Local prototypes.
224  */
225 static int	mb_ctor_mbuf(void *, int, void *, int);
226 static int	mb_ctor_clust(void *, int, void *, int);
227 static int	mb_ctor_pack(void *, int, void *, int);
228 static void	mb_dtor_mbuf(void *, int, void *);
229 static void	mb_dtor_clust(void *, int, void *);
230 static void	mb_dtor_pack(void *, int, void *);
231 static int	mb_zinit_pack(void *, int, int);
232 static void	mb_zfini_pack(void *, int);
233 
234 static void	mb_reclaim(void *);
235 static void	mbuf_init(void *);
236 static void    *mbuf_jumbo_alloc(uma_zone_t, int, u_int8_t *, int);
237 static void	mbuf_jumbo_free(void *, int, u_int8_t);
238 
239 static MALLOC_DEFINE(M_JUMBOFRAME, "jumboframes", "mbuf jumbo frame buffers");
240 
241 /* Ensure that MSIZE doesn't break dtom() - it must be a power of 2 */
242 CTASSERT((((MSIZE - 1) ^ MSIZE) + 1) >> 1 == MSIZE);
243 
244 /*
245  * Initialize FreeBSD Network buffer allocation.
246  */
247 SYSINIT(mbuf, SI_SUB_MBUF, SI_ORDER_FIRST, mbuf_init, NULL);
248 static void
249 mbuf_init(void *dummy)
250 {
251 
252 	/*
253 	 * Configure UMA zones for Mbufs, Clusters, and Packets.
254 	 */
255 	zone_mbuf = uma_zcreate(MBUF_MEM_NAME, MSIZE,
256 	    mb_ctor_mbuf, mb_dtor_mbuf,
257 #ifdef INVARIANTS
258 	    trash_init, trash_fini,
259 #else
260 	    NULL, NULL,
261 #endif
262 	    MSIZE - 1, UMA_ZONE_MAXBUCKET);
263 
264 	zone_clust = uma_zcreate(MBUF_CLUSTER_MEM_NAME, MCLBYTES,
265 	    mb_ctor_clust, mb_dtor_clust,
266 #ifdef INVARIANTS
267 	    trash_init, trash_fini,
268 #else
269 	    NULL, NULL,
270 #endif
271 	    UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
272 	if (nmbclusters > 0)
273 		uma_zone_set_max(zone_clust, nmbclusters);
274 
275 	zone_pack = uma_zsecond_create(MBUF_PACKET_MEM_NAME, mb_ctor_pack,
276 	    mb_dtor_pack, mb_zinit_pack, mb_zfini_pack, zone_mbuf);
277 
278 	/* Make jumbo frame zone too. Page size, 9k and 16k. */
279 	zone_jumbop = uma_zcreate(MBUF_JUMBOP_MEM_NAME, MJUMPAGESIZE,
280 	    mb_ctor_clust, mb_dtor_clust,
281 #ifdef INVARIANTS
282 	    trash_init, trash_fini,
283 #else
284 	    NULL, NULL,
285 #endif
286 	    UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
287 	if (nmbjumbop > 0)
288 		uma_zone_set_max(zone_jumbop, nmbjumbop);
289 
290 	zone_jumbo9 = uma_zcreate(MBUF_JUMBO9_MEM_NAME, MJUM9BYTES,
291 	    mb_ctor_clust, mb_dtor_clust,
292 #ifdef INVARIANTS
293 	    trash_init, trash_fini,
294 #else
295 	    NULL, NULL,
296 #endif
297 	    UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
298 	if (nmbjumbo9 > 0)
299 		uma_zone_set_max(zone_jumbo9, nmbjumbo9);
300 	uma_zone_set_allocf(zone_jumbo9, mbuf_jumbo_alloc);
301 	uma_zone_set_freef(zone_jumbo9, mbuf_jumbo_free);
302 
303 	zone_jumbo16 = uma_zcreate(MBUF_JUMBO16_MEM_NAME, MJUM16BYTES,
304 	    mb_ctor_clust, mb_dtor_clust,
305 #ifdef INVARIANTS
306 	    trash_init, trash_fini,
307 #else
308 	    NULL, NULL,
309 #endif
310 	    UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
311 	if (nmbjumbo16 > 0)
312 		uma_zone_set_max(zone_jumbo16, nmbjumbo16);
313 	uma_zone_set_allocf(zone_jumbo16, mbuf_jumbo_alloc);
314 	uma_zone_set_freef(zone_jumbo16, mbuf_jumbo_free);
315 
316 	zone_ext_refcnt = uma_zcreate(MBUF_EXTREFCNT_MEM_NAME, sizeof(u_int),
317 	    NULL, NULL,
318 	    NULL, NULL,
319 	    UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
320 
321 	/* uma_prealloc() goes here... */
322 
323 	/*
324 	 * Hook event handler for low-memory situation, used to
325 	 * drain protocols and push data back to the caches (UMA
326 	 * later pushes it back to VM).
327 	 */
328 	EVENTHANDLER_REGISTER(vm_lowmem, mb_reclaim, NULL,
329 	    EVENTHANDLER_PRI_FIRST);
330 
331 	/*
332 	 * [Re]set counters and local statistics knobs.
333 	 * XXX Some of these should go and be replaced, but UMA stat
334 	 * gathering needs to be revised.
335 	 */
336 	mbstat.m_mbufs = 0;
337 	mbstat.m_mclusts = 0;
338 	mbstat.m_drain = 0;
339 	mbstat.m_msize = MSIZE;
340 	mbstat.m_mclbytes = MCLBYTES;
341 	mbstat.m_minclsize = MINCLSIZE;
342 	mbstat.m_mlen = MLEN;
343 	mbstat.m_mhlen = MHLEN;
344 	mbstat.m_numtypes = MT_NTYPES;
345 
346 	mbstat.m_mcfail = mbstat.m_mpfail = 0;
347 	mbstat.sf_iocnt = 0;
348 	mbstat.sf_allocwait = mbstat.sf_allocfail = 0;
349 }
350 
351 /*
352  * UMA backend page allocator for the jumbo frame zones.
353  *
354  * Allocates kernel virtual memory that is backed by contiguous physical
355  * pages.
356  */
357 static void *
358 mbuf_jumbo_alloc(uma_zone_t zone, int bytes, u_int8_t *flags, int wait)
359 {
360 
361 	/* Inform UMA that this allocator uses kernel_map/object. */
362 	*flags = UMA_SLAB_KERNEL;
363 	return (contigmalloc(bytes, M_JUMBOFRAME, wait, (vm_paddr_t)0,
364 	    ~(vm_paddr_t)0, 1, 0));
365 }
366 
367 /*
368  * UMA backend page deallocator for the jumbo frame zones.
369  */
370 static void
371 mbuf_jumbo_free(void *mem, int size, u_int8_t flags)
372 {
373 
374 	contigfree(mem, size, M_JUMBOFRAME);
375 }
376 
377 /*
378  * Constructor for Mbuf master zone.
379  *
380  * The 'arg' pointer points to a mb_args structure which
381  * contains call-specific information required to support the
382  * mbuf allocation API.  See mbuf.h.
383  */
384 static int
385 mb_ctor_mbuf(void *mem, int size, void *arg, int how)
386 {
387 	struct mbuf *m;
388 	struct mb_args *args;
389 #ifdef MAC
390 	int error;
391 #endif
392 	int flags;
393 	short type;
394 
395 #ifdef INVARIANTS
396 	trash_ctor(mem, size, arg, how);
397 #endif
398 	m = (struct mbuf *)mem;
399 	args = (struct mb_args *)arg;
400 	flags = args->flags;
401 	type = args->type;
402 
403 	/*
404 	 * The mbuf is initialized later.  The caller has the
405 	 * responsibility to set up any MAC labels too.
406 	 */
407 	if (type == MT_NOINIT)
408 		return (0);
409 
410 	m->m_next = NULL;
411 	m->m_nextpkt = NULL;
412 	m->m_len = 0;
413 	m->m_flags = flags;
414 	m->m_type = type;
415 	if (flags & M_PKTHDR) {
416 		m->m_data = m->m_pktdat;
417 		m->m_pkthdr.rcvif = NULL;
418 		m->m_pkthdr.header = NULL;
419 		m->m_pkthdr.len = 0;
420 		m->m_pkthdr.csum_flags = 0;
421 		m->m_pkthdr.csum_data = 0;
422 		m->m_pkthdr.tso_segsz = 0;
423 		m->m_pkthdr.ether_vtag = 0;
424 		SLIST_INIT(&m->m_pkthdr.tags);
425 #ifdef MAC
426 		/* If the label init fails, fail the alloc */
427 		error = mac_mbuf_init(m, how);
428 		if (error)
429 			return (error);
430 #endif
431 	} else
432 		m->m_data = m->m_dat;
433 	return (0);
434 }
435 
436 /*
437  * The Mbuf master zone destructor.
438  */
439 static void
440 mb_dtor_mbuf(void *mem, int size, void *arg)
441 {
442 	struct mbuf *m;
443 	unsigned long flags;
444 
445 	m = (struct mbuf *)mem;
446 	flags = (unsigned long)arg;
447 
448 	if ((flags & MB_NOTAGS) == 0 && (m->m_flags & M_PKTHDR) != 0)
449 		m_tag_delete_chain(m, NULL);
450 	KASSERT((m->m_flags & M_EXT) == 0, ("%s: M_EXT set", __func__));
451 	KASSERT((m->m_flags & M_NOFREE) == 0, ("%s: M_NOFREE set", __func__));
452 #ifdef INVARIANTS
453 	trash_dtor(mem, size, arg);
454 #endif
455 }
456 
457 /*
458  * The Mbuf Packet zone destructor.
459  */
460 static void
461 mb_dtor_pack(void *mem, int size, void *arg)
462 {
463 	struct mbuf *m;
464 
465 	m = (struct mbuf *)mem;
466 	if ((m->m_flags & M_PKTHDR) != 0)
467 		m_tag_delete_chain(m, NULL);
468 
469 	/* Make sure we've got a clean cluster back. */
470 	KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__));
471 	KASSERT(m->m_ext.ext_buf != NULL, ("%s: ext_buf == NULL", __func__));
472 	KASSERT(m->m_ext.ext_free == NULL, ("%s: ext_free != NULL", __func__));
473 	KASSERT(m->m_ext.ext_arg1 == NULL, ("%s: ext_arg1 != NULL", __func__));
474 	KASSERT(m->m_ext.ext_arg2 == NULL, ("%s: ext_arg2 != NULL", __func__));
475 	KASSERT(m->m_ext.ext_size == MCLBYTES, ("%s: ext_size != MCLBYTES", __func__));
476 	KASSERT(m->m_ext.ext_type == EXT_PACKET, ("%s: ext_type != EXT_PACKET", __func__));
477 	KASSERT(*m->m_ext.ref_cnt == 1, ("%s: ref_cnt != 1", __func__));
478 #ifdef INVARIANTS
479 	trash_dtor(m->m_ext.ext_buf, MCLBYTES, arg);
480 #endif
481 	/*
482 	 * If there are processes blocked on zone_clust, waiting for pages
483 	 * to be freed up, * cause them to be woken up by draining the
484 	 * packet zone.  We are exposed to a race here * (in the check for
485 	 * the UMA_ZFLAG_FULL) where we might miss the flag set, but that
486 	 * is deliberate. We don't want to acquire the zone lock for every
487 	 * mbuf free.
488 	 */
489 	if (uma_zone_exhausted_nolock(zone_clust))
490 		zone_drain(zone_pack);
491 }
492 
493 /*
494  * The Cluster and Jumbo[PAGESIZE|9|16] zone constructor.
495  *
496  * Here the 'arg' pointer points to the Mbuf which we
497  * are configuring cluster storage for.  If 'arg' is
498  * empty we allocate just the cluster without setting
499  * the mbuf to it.  See mbuf.h.
500  */
501 static int
502 mb_ctor_clust(void *mem, int size, void *arg, int how)
503 {
504 	struct mbuf *m;
505 	u_int *refcnt;
506 	int type;
507 	uma_zone_t zone;
508 
509 #ifdef INVARIANTS
510 	trash_ctor(mem, size, arg, how);
511 #endif
512 	switch (size) {
513 	case MCLBYTES:
514 		type = EXT_CLUSTER;
515 		zone = zone_clust;
516 		break;
517 #if MJUMPAGESIZE != MCLBYTES
518 	case MJUMPAGESIZE:
519 		type = EXT_JUMBOP;
520 		zone = zone_jumbop;
521 		break;
522 #endif
523 	case MJUM9BYTES:
524 		type = EXT_JUMBO9;
525 		zone = zone_jumbo9;
526 		break;
527 	case MJUM16BYTES:
528 		type = EXT_JUMBO16;
529 		zone = zone_jumbo16;
530 		break;
531 	default:
532 		panic("unknown cluster size");
533 		break;
534 	}
535 
536 	m = (struct mbuf *)arg;
537 	refcnt = uma_find_refcnt(zone, mem);
538 	*refcnt = 1;
539 	if (m != NULL) {
540 		m->m_ext.ext_buf = (caddr_t)mem;
541 		m->m_data = m->m_ext.ext_buf;
542 		m->m_flags |= M_EXT;
543 		m->m_ext.ext_free = NULL;
544 		m->m_ext.ext_arg1 = NULL;
545 		m->m_ext.ext_arg2 = NULL;
546 		m->m_ext.ext_size = size;
547 		m->m_ext.ext_type = type;
548 		m->m_ext.ref_cnt = refcnt;
549 	}
550 
551 	return (0);
552 }
553 
554 /*
555  * The Mbuf Cluster zone destructor.
556  */
557 static void
558 mb_dtor_clust(void *mem, int size, void *arg)
559 {
560 #ifdef INVARIANTS
561 	uma_zone_t zone;
562 
563 	zone = m_getzone(size);
564 	KASSERT(*(uma_find_refcnt(zone, mem)) <= 1,
565 		("%s: refcnt incorrect %u", __func__,
566 		 *(uma_find_refcnt(zone, mem))) );
567 
568 	trash_dtor(mem, size, arg);
569 #endif
570 }
571 
572 /*
573  * The Packet secondary zone's init routine, executed on the
574  * object's transition from mbuf keg slab to zone cache.
575  */
576 static int
577 mb_zinit_pack(void *mem, int size, int how)
578 {
579 	struct mbuf *m;
580 
581 	m = (struct mbuf *)mem;		/* m is virgin. */
582 	if (uma_zalloc_arg(zone_clust, m, how) == NULL ||
583 	    m->m_ext.ext_buf == NULL)
584 		return (ENOMEM);
585 	m->m_ext.ext_type = EXT_PACKET;	/* Override. */
586 #ifdef INVARIANTS
587 	trash_init(m->m_ext.ext_buf, MCLBYTES, how);
588 #endif
589 	return (0);
590 }
591 
592 /*
593  * The Packet secondary zone's fini routine, executed on the
594  * object's transition from zone cache to keg slab.
595  */
596 static void
597 mb_zfini_pack(void *mem, int size)
598 {
599 	struct mbuf *m;
600 
601 	m = (struct mbuf *)mem;
602 #ifdef INVARIANTS
603 	trash_fini(m->m_ext.ext_buf, MCLBYTES);
604 #endif
605 	uma_zfree_arg(zone_clust, m->m_ext.ext_buf, NULL);
606 #ifdef INVARIANTS
607 	trash_dtor(mem, size, NULL);
608 #endif
609 }
610 
611 /*
612  * The "packet" keg constructor.
613  */
614 static int
615 mb_ctor_pack(void *mem, int size, void *arg, int how)
616 {
617 	struct mbuf *m;
618 	struct mb_args *args;
619 #ifdef MAC
620 	int error;
621 #endif
622 	int flags;
623 	short type;
624 
625 	m = (struct mbuf *)mem;
626 	args = (struct mb_args *)arg;
627 	flags = args->flags;
628 	type = args->type;
629 
630 #ifdef INVARIANTS
631 	trash_ctor(m->m_ext.ext_buf, MCLBYTES, arg, how);
632 #endif
633 	m->m_next = NULL;
634 	m->m_nextpkt = NULL;
635 	m->m_data = m->m_ext.ext_buf;
636 	m->m_len = 0;
637 	m->m_flags = (flags | M_EXT);
638 	m->m_type = type;
639 
640 	if (flags & M_PKTHDR) {
641 		m->m_pkthdr.rcvif = NULL;
642 		m->m_pkthdr.len = 0;
643 		m->m_pkthdr.header = NULL;
644 		m->m_pkthdr.csum_flags = 0;
645 		m->m_pkthdr.csum_data = 0;
646 		m->m_pkthdr.tso_segsz = 0;
647 		m->m_pkthdr.ether_vtag = 0;
648 		SLIST_INIT(&m->m_pkthdr.tags);
649 #ifdef MAC
650 		/* If the label init fails, fail the alloc */
651 		error = mac_mbuf_init(m, how);
652 		if (error)
653 			return (error);
654 #endif
655 	}
656 	/* m_ext is already initialized. */
657 
658 	return (0);
659 }
660 
661 /*
662  * This is the protocol drain routine.
663  *
664  * No locks should be held when this is called.  The drain routines have to
665  * presently acquire some locks which raises the possibility of lock order
666  * reversal.
667  */
668 static void
669 mb_reclaim(void *junk)
670 {
671 	struct domain *dp;
672 	struct protosw *pr;
673 
674 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK | WARN_PANIC, NULL,
675 	    "mb_reclaim()");
676 
677 	for (dp = domains; dp != NULL; dp = dp->dom_next)
678 		for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++)
679 			if (pr->pr_drain != NULL)
680 				(*pr->pr_drain)();
681 }
682