xref: /freebsd/sys/kern/kern_mbuf.c (revision a10cee30c94cf5944826d2a495e9cdf339dfbcc8)
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_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 
99 int nmbclusters;		/* limits number of mbuf clusters */
100 int nmbjumbop;			/* limits number of page size jumbo clusters */
101 int nmbjumbo9;			/* limits number of 9k jumbo clusters */
102 int nmbjumbo16;			/* limits number of 16k jumbo clusters */
103 struct mbstat mbstat;
104 
105 /*
106  * tunable_mbinit() has to be run before init_maxsockets() thus
107  * the SYSINIT order below is SI_ORDER_MIDDLE while init_maxsockets()
108  * runs at SI_ORDER_ANY.
109  */
110 static void
111 tunable_mbinit(void *dummy)
112 {
113 
114 	/* This has to be done before VM init. */
115 	TUNABLE_INT_FETCH("kern.ipc.nmbclusters", &nmbclusters);
116 	if (nmbclusters == 0)
117 		nmbclusters = 1024 + maxusers * 64;
118 
119 	TUNABLE_INT_FETCH("kern.ipc.nmbjumbop", &nmbjumbop);
120 	if (nmbjumbop == 0)
121 		nmbjumbop = nmbclusters / 2;
122 
123 	TUNABLE_INT_FETCH("kern.ipc.nmbjumbo9", &nmbjumbo9);
124 	if (nmbjumbo9 == 0)
125 		nmbjumbo9 = nmbclusters / 4;
126 
127 	TUNABLE_INT_FETCH("kern.ipc.nmbjumbo16", &nmbjumbo16);
128 	if (nmbjumbo16 == 0)
129 		nmbjumbo16 = nmbclusters / 8;
130 }
131 SYSINIT(tunable_mbinit, SI_SUB_TUNABLES, SI_ORDER_MIDDLE, tunable_mbinit, NULL);
132 
133 static int
134 sysctl_nmbclusters(SYSCTL_HANDLER_ARGS)
135 {
136 	int error, newnmbclusters;
137 
138 	newnmbclusters = nmbclusters;
139 	error = sysctl_handle_int(oidp, &newnmbclusters, 0, req);
140 	if (error == 0 && req->newptr) {
141 		if (newnmbclusters > nmbclusters) {
142 			nmbclusters = newnmbclusters;
143 			uma_zone_set_max(zone_clust, nmbclusters);
144 			EVENTHANDLER_INVOKE(nmbclusters_change);
145 		} else
146 			error = EINVAL;
147 	}
148 	return (error);
149 }
150 SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbclusters, CTLTYPE_INT|CTLFLAG_RW,
151 &nmbclusters, 0, sysctl_nmbclusters, "IU",
152     "Maximum number of mbuf clusters allowed");
153 
154 static int
155 sysctl_nmbjumbop(SYSCTL_HANDLER_ARGS)
156 {
157 	int error, newnmbjumbop;
158 
159 	newnmbjumbop = nmbjumbop;
160 	error = sysctl_handle_int(oidp, &newnmbjumbop, 0, req);
161 	if (error == 0 && req->newptr) {
162 		if (newnmbjumbop> nmbjumbop) {
163 			nmbjumbop = newnmbjumbop;
164 			uma_zone_set_max(zone_jumbop, nmbjumbop);
165 		} else
166 			error = EINVAL;
167 	}
168 	return (error);
169 }
170 SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbop, CTLTYPE_INT|CTLFLAG_RW,
171 &nmbjumbop, 0, sysctl_nmbjumbop, "IU",
172 	 "Maximum number of mbuf page size jumbo clusters allowed");
173 
174 
175 static int
176 sysctl_nmbjumbo9(SYSCTL_HANDLER_ARGS)
177 {
178 	int error, newnmbjumbo9;
179 
180 	newnmbjumbo9 = nmbjumbo9;
181 	error = sysctl_handle_int(oidp, &newnmbjumbo9, 0, req);
182 	if (error == 0 && req->newptr) {
183 		if (newnmbjumbo9> nmbjumbo9) {
184 			nmbjumbo9 = newnmbjumbo9;
185 			uma_zone_set_max(zone_jumbo9, nmbjumbo9);
186 		} else
187 			error = EINVAL;
188 	}
189 	return (error);
190 }
191 SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbo9, CTLTYPE_INT|CTLFLAG_RW,
192 &nmbjumbo9, 0, sysctl_nmbjumbo9, "IU",
193 	"Maximum number of mbuf 9k jumbo clusters allowed");
194 
195 static int
196 sysctl_nmbjumbo16(SYSCTL_HANDLER_ARGS)
197 {
198 	int error, newnmbjumbo16;
199 
200 	newnmbjumbo16 = nmbjumbo16;
201 	error = sysctl_handle_int(oidp, &newnmbjumbo16, 0, req);
202 	if (error == 0 && req->newptr) {
203 		if (newnmbjumbo16> nmbjumbo16) {
204 			nmbjumbo16 = newnmbjumbo16;
205 			uma_zone_set_max(zone_jumbo16, nmbjumbo16);
206 		} else
207 			error = EINVAL;
208 	}
209 	return (error);
210 }
211 SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbo16, CTLTYPE_INT|CTLFLAG_RW,
212 &nmbjumbo16, 0, sysctl_nmbjumbo16, "IU",
213     "Maximum number of mbuf 16k jumbo clusters allowed");
214 
215 
216 
217 SYSCTL_STRUCT(_kern_ipc, OID_AUTO, mbstat, CTLFLAG_RD, &mbstat, mbstat,
218     "Mbuf general information and statistics");
219 
220 /*
221  * Zones from which we allocate.
222  */
223 uma_zone_t	zone_mbuf;
224 uma_zone_t	zone_clust;
225 uma_zone_t	zone_pack;
226 uma_zone_t	zone_jumbop;
227 uma_zone_t	zone_jumbo9;
228 uma_zone_t	zone_jumbo16;
229 uma_zone_t	zone_ext_refcnt;
230 
231 /*
232  * Local prototypes.
233  */
234 static int	mb_ctor_mbuf(void *, int, void *, int);
235 static int	mb_ctor_clust(void *, int, void *, int);
236 static int	mb_ctor_pack(void *, int, void *, int);
237 static void	mb_dtor_mbuf(void *, int, void *);
238 static void	mb_dtor_clust(void *, int, void *);
239 static void	mb_dtor_pack(void *, int, void *);
240 static int	mb_zinit_pack(void *, int, int);
241 static void	mb_zfini_pack(void *, int);
242 
243 static void	mb_reclaim(void *);
244 static void	mbuf_init(void *);
245 static void    *mbuf_jumbo_alloc(uma_zone_t, int, uint8_t *, int);
246 
247 /* Ensure that MSIZE doesn't break dtom() - it must be a power of 2 */
248 CTASSERT((((MSIZE - 1) ^ MSIZE) + 1) >> 1 == MSIZE);
249 
250 /*
251  * Initialize FreeBSD Network buffer allocation.
252  */
253 SYSINIT(mbuf, SI_SUB_MBUF, SI_ORDER_FIRST, mbuf_init, NULL);
254 static void
255 mbuf_init(void *dummy)
256 {
257 
258 	/*
259 	 * Configure UMA zones for Mbufs, Clusters, and Packets.
260 	 */
261 	zone_mbuf = uma_zcreate(MBUF_MEM_NAME, MSIZE,
262 	    mb_ctor_mbuf, mb_dtor_mbuf,
263 #ifdef INVARIANTS
264 	    trash_init, trash_fini,
265 #else
266 	    NULL, NULL,
267 #endif
268 	    MSIZE - 1, UMA_ZONE_MAXBUCKET);
269 
270 	zone_clust = uma_zcreate(MBUF_CLUSTER_MEM_NAME, MCLBYTES,
271 	    mb_ctor_clust, mb_dtor_clust,
272 #ifdef INVARIANTS
273 	    trash_init, trash_fini,
274 #else
275 	    NULL, NULL,
276 #endif
277 	    UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
278 	if (nmbclusters > 0)
279 		uma_zone_set_max(zone_clust, nmbclusters);
280 
281 	zone_pack = uma_zsecond_create(MBUF_PACKET_MEM_NAME, mb_ctor_pack,
282 	    mb_dtor_pack, mb_zinit_pack, mb_zfini_pack, zone_mbuf);
283 
284 	/* Make jumbo frame zone too. Page size, 9k and 16k. */
285 	zone_jumbop = uma_zcreate(MBUF_JUMBOP_MEM_NAME, MJUMPAGESIZE,
286 	    mb_ctor_clust, mb_dtor_clust,
287 #ifdef INVARIANTS
288 	    trash_init, trash_fini,
289 #else
290 	    NULL, NULL,
291 #endif
292 	    UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
293 	if (nmbjumbop > 0)
294 		uma_zone_set_max(zone_jumbop, nmbjumbop);
295 
296 	zone_jumbo9 = uma_zcreate(MBUF_JUMBO9_MEM_NAME, MJUM9BYTES,
297 	    mb_ctor_clust, mb_dtor_clust,
298 #ifdef INVARIANTS
299 	    trash_init, trash_fini,
300 #else
301 	    NULL, NULL,
302 #endif
303 	    UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
304 	if (nmbjumbo9 > 0)
305 		uma_zone_set_max(zone_jumbo9, nmbjumbo9);
306 	uma_zone_set_allocf(zone_jumbo9, mbuf_jumbo_alloc);
307 
308 	zone_jumbo16 = uma_zcreate(MBUF_JUMBO16_MEM_NAME, MJUM16BYTES,
309 	    mb_ctor_clust, mb_dtor_clust,
310 #ifdef INVARIANTS
311 	    trash_init, trash_fini,
312 #else
313 	    NULL, NULL,
314 #endif
315 	    UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
316 	if (nmbjumbo16 > 0)
317 		uma_zone_set_max(zone_jumbo16, nmbjumbo16);
318 	uma_zone_set_allocf(zone_jumbo16, mbuf_jumbo_alloc);
319 
320 	zone_ext_refcnt = uma_zcreate(MBUF_EXTREFCNT_MEM_NAME, sizeof(u_int),
321 	    NULL, NULL,
322 	    NULL, NULL,
323 	    UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
324 
325 	/* uma_prealloc() goes here... */
326 
327 	/*
328 	 * Hook event handler for low-memory situation, used to
329 	 * drain protocols and push data back to the caches (UMA
330 	 * later pushes it back to VM).
331 	 */
332 	EVENTHANDLER_REGISTER(vm_lowmem, mb_reclaim, NULL,
333 	    EVENTHANDLER_PRI_FIRST);
334 
335 	/*
336 	 * [Re]set counters and local statistics knobs.
337 	 * XXX Some of these should go and be replaced, but UMA stat
338 	 * gathering needs to be revised.
339 	 */
340 	mbstat.m_mbufs = 0;
341 	mbstat.m_mclusts = 0;
342 	mbstat.m_drain = 0;
343 	mbstat.m_msize = MSIZE;
344 	mbstat.m_mclbytes = MCLBYTES;
345 	mbstat.m_minclsize = MINCLSIZE;
346 	mbstat.m_mlen = MLEN;
347 	mbstat.m_mhlen = MHLEN;
348 	mbstat.m_numtypes = MT_NTYPES;
349 
350 	mbstat.m_mcfail = mbstat.m_mpfail = 0;
351 	mbstat.sf_iocnt = 0;
352 	mbstat.sf_allocwait = mbstat.sf_allocfail = 0;
353 }
354 
355 /*
356  * UMA backend page allocator for the jumbo frame zones.
357  *
358  * Allocates kernel virtual memory that is backed by contiguous physical
359  * pages.
360  */
361 static void *
362 mbuf_jumbo_alloc(uma_zone_t zone, int bytes, uint8_t *flags, int wait)
363 {
364 
365 	/* Inform UMA that this allocator uses kernel_map/object. */
366 	*flags = UMA_SLAB_KERNEL;
367 	return ((void *)kmem_alloc_contig(kernel_map, bytes, wait,
368 	    (vm_paddr_t)0, ~(vm_paddr_t)0, 1, 0, VM_MEMATTR_DEFAULT));
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.header = NULL;
413 		m->m_pkthdr.len = 0;
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 		m->m_pkthdr.flowid = 0;
419 		SLIST_INIT(&m->m_pkthdr.tags);
420 #ifdef MAC
421 		/* If the label init fails, fail the alloc */
422 		error = mac_mbuf_init(m, how);
423 		if (error)
424 			return (error);
425 #endif
426 	} else
427 		m->m_data = m->m_dat;
428 	return (0);
429 }
430 
431 /*
432  * The Mbuf master zone destructor.
433  */
434 static void
435 mb_dtor_mbuf(void *mem, int size, void *arg)
436 {
437 	struct mbuf *m;
438 	unsigned long flags;
439 
440 	m = (struct mbuf *)mem;
441 	flags = (unsigned long)arg;
442 
443 	if ((flags & MB_NOTAGS) == 0 && (m->m_flags & M_PKTHDR) != 0)
444 		m_tag_delete_chain(m, NULL);
445 	KASSERT((m->m_flags & M_EXT) == 0, ("%s: M_EXT set", __func__));
446 	KASSERT((m->m_flags & M_NOFREE) == 0, ("%s: M_NOFREE set", __func__));
447 #ifdef INVARIANTS
448 	trash_dtor(mem, size, arg);
449 #endif
450 }
451 
452 /*
453  * The Mbuf Packet zone destructor.
454  */
455 static void
456 mb_dtor_pack(void *mem, int size, void *arg)
457 {
458 	struct mbuf *m;
459 
460 	m = (struct mbuf *)mem;
461 	if ((m->m_flags & M_PKTHDR) != 0)
462 		m_tag_delete_chain(m, NULL);
463 
464 	/* Make sure we've got a clean cluster back. */
465 	KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__));
466 	KASSERT(m->m_ext.ext_buf != NULL, ("%s: ext_buf == NULL", __func__));
467 	KASSERT(m->m_ext.ext_free == NULL, ("%s: ext_free != NULL", __func__));
468 	KASSERT(m->m_ext.ext_arg1 == NULL, ("%s: ext_arg1 != NULL", __func__));
469 	KASSERT(m->m_ext.ext_arg2 == NULL, ("%s: ext_arg2 != NULL", __func__));
470 	KASSERT(m->m_ext.ext_size == MCLBYTES, ("%s: ext_size != MCLBYTES", __func__));
471 	KASSERT(m->m_ext.ext_type == EXT_PACKET, ("%s: ext_type != EXT_PACKET", __func__));
472 	KASSERT(*m->m_ext.ref_cnt == 1, ("%s: ref_cnt != 1", __func__));
473 #ifdef INVARIANTS
474 	trash_dtor(m->m_ext.ext_buf, MCLBYTES, arg);
475 #endif
476 	/*
477 	 * If there are processes blocked on zone_clust, waiting for pages
478 	 * to be freed up, * cause them to be woken up by draining the
479 	 * packet zone.  We are exposed to a race here * (in the check for
480 	 * the UMA_ZFLAG_FULL) where we might miss the flag set, but that
481 	 * is deliberate. We don't want to acquire the zone lock for every
482 	 * mbuf free.
483 	 */
484 	if (uma_zone_exhausted_nolock(zone_clust))
485 		zone_drain(zone_pack);
486 }
487 
488 /*
489  * The Cluster and Jumbo[PAGESIZE|9|16] zone constructor.
490  *
491  * Here the 'arg' pointer points to the Mbuf which we
492  * are configuring cluster storage for.  If 'arg' is
493  * empty we allocate just the cluster without setting
494  * the mbuf to it.  See mbuf.h.
495  */
496 static int
497 mb_ctor_clust(void *mem, int size, void *arg, int how)
498 {
499 	struct mbuf *m;
500 	u_int *refcnt;
501 	int type;
502 	uma_zone_t zone;
503 
504 #ifdef INVARIANTS
505 	trash_ctor(mem, size, arg, how);
506 #endif
507 	switch (size) {
508 	case MCLBYTES:
509 		type = EXT_CLUSTER;
510 		zone = zone_clust;
511 		break;
512 #if MJUMPAGESIZE != MCLBYTES
513 	case MJUMPAGESIZE:
514 		type = EXT_JUMBOP;
515 		zone = zone_jumbop;
516 		break;
517 #endif
518 	case MJUM9BYTES:
519 		type = EXT_JUMBO9;
520 		zone = zone_jumbo9;
521 		break;
522 	case MJUM16BYTES:
523 		type = EXT_JUMBO16;
524 		zone = zone_jumbo16;
525 		break;
526 	default:
527 		panic("unknown cluster size");
528 		break;
529 	}
530 
531 	m = (struct mbuf *)arg;
532 	refcnt = uma_find_refcnt(zone, mem);
533 	*refcnt = 1;
534 	if (m != NULL) {
535 		m->m_ext.ext_buf = (caddr_t)mem;
536 		m->m_data = m->m_ext.ext_buf;
537 		m->m_flags |= M_EXT;
538 		m->m_ext.ext_free = NULL;
539 		m->m_ext.ext_arg1 = NULL;
540 		m->m_ext.ext_arg2 = NULL;
541 		m->m_ext.ext_size = size;
542 		m->m_ext.ext_type = type;
543 		m->m_ext.ref_cnt = refcnt;
544 	}
545 
546 	return (0);
547 }
548 
549 /*
550  * The Mbuf Cluster zone destructor.
551  */
552 static void
553 mb_dtor_clust(void *mem, int size, void *arg)
554 {
555 #ifdef INVARIANTS
556 	uma_zone_t zone;
557 
558 	zone = m_getzone(size);
559 	KASSERT(*(uma_find_refcnt(zone, mem)) <= 1,
560 		("%s: refcnt incorrect %u", __func__,
561 		 *(uma_find_refcnt(zone, mem))) );
562 
563 	trash_dtor(mem, size, arg);
564 #endif
565 }
566 
567 /*
568  * The Packet secondary zone's init routine, executed on the
569  * object's transition from mbuf keg slab to zone cache.
570  */
571 static int
572 mb_zinit_pack(void *mem, int size, int how)
573 {
574 	struct mbuf *m;
575 
576 	m = (struct mbuf *)mem;		/* m is virgin. */
577 	if (uma_zalloc_arg(zone_clust, m, how) == NULL ||
578 	    m->m_ext.ext_buf == NULL)
579 		return (ENOMEM);
580 	m->m_ext.ext_type = EXT_PACKET;	/* Override. */
581 #ifdef INVARIANTS
582 	trash_init(m->m_ext.ext_buf, MCLBYTES, how);
583 #endif
584 	return (0);
585 }
586 
587 /*
588  * The Packet secondary zone's fini routine, executed on the
589  * object's transition from zone cache to keg slab.
590  */
591 static void
592 mb_zfini_pack(void *mem, int size)
593 {
594 	struct mbuf *m;
595 
596 	m = (struct mbuf *)mem;
597 #ifdef INVARIANTS
598 	trash_fini(m->m_ext.ext_buf, MCLBYTES);
599 #endif
600 	uma_zfree_arg(zone_clust, m->m_ext.ext_buf, NULL);
601 #ifdef INVARIANTS
602 	trash_dtor(mem, size, NULL);
603 #endif
604 }
605 
606 /*
607  * The "packet" keg constructor.
608  */
609 static int
610 mb_ctor_pack(void *mem, int size, void *arg, int how)
611 {
612 	struct mbuf *m;
613 	struct mb_args *args;
614 #ifdef MAC
615 	int error;
616 #endif
617 	int flags;
618 	short type;
619 
620 	m = (struct mbuf *)mem;
621 	args = (struct mb_args *)arg;
622 	flags = args->flags;
623 	type = args->type;
624 
625 #ifdef INVARIANTS
626 	trash_ctor(m->m_ext.ext_buf, MCLBYTES, arg, how);
627 #endif
628 	m->m_next = NULL;
629 	m->m_nextpkt = NULL;
630 	m->m_data = m->m_ext.ext_buf;
631 	m->m_len = 0;
632 	m->m_flags = (flags | M_EXT);
633 	m->m_type = type;
634 
635 	if (flags & M_PKTHDR) {
636 		m->m_pkthdr.rcvif = NULL;
637 		m->m_pkthdr.len = 0;
638 		m->m_pkthdr.header = NULL;
639 		m->m_pkthdr.csum_flags = 0;
640 		m->m_pkthdr.csum_data = 0;
641 		m->m_pkthdr.tso_segsz = 0;
642 		m->m_pkthdr.ether_vtag = 0;
643 		m->m_pkthdr.flowid = 0;
644 		SLIST_INIT(&m->m_pkthdr.tags);
645 #ifdef MAC
646 		/* If the label init fails, fail the alloc */
647 		error = mac_mbuf_init(m, how);
648 		if (error)
649 			return (error);
650 #endif
651 	}
652 	/* m_ext is already initialized. */
653 
654 	return (0);
655 }
656 
657 int
658 m_pkthdr_init(struct mbuf *m, int how)
659 {
660 #ifdef MAC
661 	int error;
662 #endif
663 	m->m_data = m->m_pktdat;
664 	SLIST_INIT(&m->m_pkthdr.tags);
665 	m->m_pkthdr.rcvif = NULL;
666 	m->m_pkthdr.header = NULL;
667 	m->m_pkthdr.len = 0;
668 	m->m_pkthdr.flowid = 0;
669 	m->m_pkthdr.csum_flags = 0;
670 	m->m_pkthdr.csum_data = 0;
671 	m->m_pkthdr.tso_segsz = 0;
672 	m->m_pkthdr.ether_vtag = 0;
673 #ifdef MAC
674 	/* If the label init fails, fail the alloc */
675 	error = mac_mbuf_init(m, how);
676 	if (error)
677 		return (error);
678 #endif
679 
680 	return (0);
681 }
682 
683 /*
684  * This is the protocol drain routine.
685  *
686  * No locks should be held when this is called.  The drain routines have to
687  * presently acquire some locks which raises the possibility of lock order
688  * reversal.
689  */
690 static void
691 mb_reclaim(void *junk)
692 {
693 	struct domain *dp;
694 	struct protosw *pr;
695 
696 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK | WARN_PANIC, NULL,
697 	    "mb_reclaim()");
698 
699 	for (dp = domains; dp != NULL; dp = dp->dom_next)
700 		for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++)
701 			if (pr->pr_drain != NULL)
702 				(*pr->pr_drain)();
703 }
704