xref: /freebsd/sys/kern/kern_mbuf.c (revision d5fc25e5d6c52b306312784663ccad85923a9c76)
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_page.h>
48 #include <vm/uma.h>
49 #include <vm/uma_int.h>
50 #include <vm/uma_dbg.h>
51 
52 /*
53  * In FreeBSD, Mbufs and Mbuf Clusters are allocated from UMA
54  * Zones.
55  *
56  * Mbuf Clusters (2K, contiguous) are allocated from the Cluster
57  * Zone.  The Zone can be capped at kern.ipc.nmbclusters, if the
58  * administrator so desires.
59  *
60  * Mbufs are allocated from a UMA Master Zone called the Mbuf
61  * Zone.
62  *
63  * Additionally, FreeBSD provides a Packet Zone, which it
64  * configures as a Secondary Zone to the Mbuf Master Zone,
65  * thus sharing backend Slab kegs with the Mbuf Master Zone.
66  *
67  * Thus common-case allocations and locking are simplified:
68  *
69  *  m_clget()                m_getcl()
70  *    |                         |
71  *    |   .------------>[(Packet Cache)]    m_get(), m_gethdr()
72  *    |   |             [     Packet   ]            |
73  *  [(Cluster Cache)]   [    Secondary ]   [ (Mbuf Cache)     ]
74  *  [ Cluster Zone  ]   [     Zone     ]   [ Mbuf Master Zone ]
75  *        |                       \________         |
76  *  [ Cluster Keg   ]                      \       /
77  *        |    	                         [ Mbuf Keg   ]
78  *  [ Cluster Slabs ]                         |
79  *        |                              [ Mbuf Slabs ]
80  *         \____________(VM)_________________/
81  *
82  *
83  * Whenever an object is allocated with uma_zalloc() out of
84  * one of the Zones its _ctor_ function is executed.  The same
85  * for any deallocation through uma_zfree() the _dtor_ function
86  * is executed.
87  *
88  * Caches are per-CPU and are filled from the Master Zone.
89  *
90  * Whenever an object is allocated from the underlying global
91  * memory pool it gets pre-initialized with the _zinit_ functions.
92  * When the Keg's are overfull objects get decomissioned with
93  * _zfini_ functions and free'd back to the global memory pool.
94  *
95  */
96 
97 int nmbclusters;		/* limits number of mbuf clusters */
98 int nmbjumbop;			/* limits number of page size jumbo clusters */
99 int nmbjumbo9;			/* limits number of 9k jumbo clusters */
100 int nmbjumbo16;			/* limits number of 16k jumbo clusters */
101 struct mbstat mbstat;
102 
103 /*
104  * tunable_mbinit() has to be run before init_maxsockets() thus
105  * the SYSINIT order below is SI_ORDER_MIDDLE while init_maxsockets()
106  * runs at SI_ORDER_ANY.
107  */
108 static void
109 tunable_mbinit(void *dummy)
110 {
111 	TUNABLE_INT_FETCH("kern.ipc.nmbclusters", &nmbclusters);
112 
113 	/* This has to be done before VM init. */
114 	if (nmbclusters == 0)
115 		nmbclusters = 1024 + maxusers * 64;
116 	nmbjumbop = nmbclusters / 2;
117 	nmbjumbo9 = nmbjumbop / 2;
118 	nmbjumbo16 = nmbjumbo9 / 2;
119 }
120 SYSINIT(tunable_mbinit, SI_SUB_TUNABLES, SI_ORDER_MIDDLE, tunable_mbinit, NULL);
121 
122 static int
123 sysctl_nmbclusters(SYSCTL_HANDLER_ARGS)
124 {
125 	int error, newnmbclusters;
126 
127 	newnmbclusters = nmbclusters;
128 	error = sysctl_handle_int(oidp, &newnmbclusters, 0, req);
129 	if (error == 0 && req->newptr) {
130 		if (newnmbclusters > nmbclusters) {
131 			nmbclusters = newnmbclusters;
132 			uma_zone_set_max(zone_clust, nmbclusters);
133 			EVENTHANDLER_INVOKE(nmbclusters_change);
134 		} else
135 			error = EINVAL;
136 	}
137 	return (error);
138 }
139 SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbclusters, CTLTYPE_INT|CTLFLAG_RW,
140 &nmbclusters, 0, sysctl_nmbclusters, "IU",
141     "Maximum number of mbuf clusters allowed");
142 
143 static int
144 sysctl_nmbjumbop(SYSCTL_HANDLER_ARGS)
145 {
146 	int error, newnmbjumbop;
147 
148 	newnmbjumbop = nmbjumbop;
149 	error = sysctl_handle_int(oidp, &newnmbjumbop, 0, req);
150 	if (error == 0 && req->newptr) {
151 		if (newnmbjumbop> nmbjumbop) {
152 			nmbjumbop = newnmbjumbop;
153 			uma_zone_set_max(zone_jumbop, nmbjumbop);
154 		} else
155 			error = EINVAL;
156 	}
157 	return (error);
158 }
159 SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbop, CTLTYPE_INT|CTLFLAG_RW,
160 &nmbjumbop, 0, sysctl_nmbjumbop, "IU",
161 	 "Maximum number of mbuf page size jumbo clusters allowed");
162 
163 
164 static int
165 sysctl_nmbjumbo9(SYSCTL_HANDLER_ARGS)
166 {
167 	int error, newnmbjumbo9;
168 
169 	newnmbjumbo9 = nmbjumbo9;
170 	error = sysctl_handle_int(oidp, &newnmbjumbo9, 0, req);
171 	if (error == 0 && req->newptr) {
172 		if (newnmbjumbo9> nmbjumbo9) {
173 			nmbjumbo9 = newnmbjumbo9;
174 			uma_zone_set_max(zone_jumbo9, nmbjumbo9);
175 		} else
176 			error = EINVAL;
177 	}
178 	return (error);
179 }
180 SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbo9, CTLTYPE_INT|CTLFLAG_RW,
181 &nmbjumbo9, 0, sysctl_nmbjumbo9, "IU",
182 	"Maximum number of mbuf 9k jumbo clusters allowed");
183 
184 static int
185 sysctl_nmbjumbo16(SYSCTL_HANDLER_ARGS)
186 {
187 	int error, newnmbjumbo16;
188 
189 	newnmbjumbo16 = nmbjumbo16;
190 	error = sysctl_handle_int(oidp, &newnmbjumbo16, 0, req);
191 	if (error == 0 && req->newptr) {
192 		if (newnmbjumbo16> nmbjumbo16) {
193 			nmbjumbo16 = newnmbjumbo16;
194 			uma_zone_set_max(zone_jumbo16, nmbjumbo16);
195 		} else
196 			error = EINVAL;
197 	}
198 	return (error);
199 }
200 SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbo16, CTLTYPE_INT|CTLFLAG_RW,
201 &nmbjumbo16, 0, sysctl_nmbjumbo16, "IU",
202     "Maximum number of mbuf 16k jumbo clusters allowed");
203 
204 
205 
206 SYSCTL_STRUCT(_kern_ipc, OID_AUTO, mbstat, CTLFLAG_RD, &mbstat, mbstat,
207     "Mbuf general information and statistics");
208 
209 /*
210  * Zones from which we allocate.
211  */
212 uma_zone_t	zone_mbuf;
213 uma_zone_t	zone_clust;
214 uma_zone_t	zone_pack;
215 uma_zone_t	zone_jumbop;
216 uma_zone_t	zone_jumbo9;
217 uma_zone_t	zone_jumbo16;
218 uma_zone_t	zone_ext_refcnt;
219 
220 /*
221  * Local prototypes.
222  */
223 static int	mb_ctor_mbuf(void *, int, void *, int);
224 static int	mb_ctor_clust(void *, int, void *, int);
225 static int	mb_ctor_pack(void *, int, void *, int);
226 static void	mb_dtor_mbuf(void *, int, void *);
227 static void	mb_dtor_clust(void *, int, void *);
228 static void	mb_dtor_pack(void *, int, void *);
229 static int	mb_zinit_pack(void *, int, int);
230 static void	mb_zfini_pack(void *, int);
231 
232 static void	mb_reclaim(void *);
233 static void	mbuf_init(void *);
234 static void    *mbuf_jumbo_alloc(uma_zone_t, int, u_int8_t *, int);
235 static void	mbuf_jumbo_free(void *, int, u_int8_t);
236 
237 static MALLOC_DEFINE(M_JUMBOFRAME, "jumboframes", "mbuf jumbo frame buffers");
238 
239 /* Ensure that MSIZE doesn't break dtom() - it must be a power of 2 */
240 CTASSERT((((MSIZE - 1) ^ MSIZE) + 1) >> 1 == MSIZE);
241 
242 /*
243  * Initialize FreeBSD Network buffer allocation.
244  */
245 SYSINIT(mbuf, SI_SUB_MBUF, SI_ORDER_FIRST, mbuf_init, NULL);
246 static void
247 mbuf_init(void *dummy)
248 {
249 
250 	/*
251 	 * Configure UMA zones for Mbufs, Clusters, and Packets.
252 	 */
253 	zone_mbuf = uma_zcreate(MBUF_MEM_NAME, MSIZE,
254 	    mb_ctor_mbuf, mb_dtor_mbuf,
255 #ifdef INVARIANTS
256 	    trash_init, trash_fini,
257 #else
258 	    NULL, NULL,
259 #endif
260 	    MSIZE - 1, UMA_ZONE_MAXBUCKET);
261 
262 	zone_clust = uma_zcreate(MBUF_CLUSTER_MEM_NAME, MCLBYTES,
263 	    mb_ctor_clust, mb_dtor_clust,
264 #ifdef INVARIANTS
265 	    trash_init, trash_fini,
266 #else
267 	    NULL, NULL,
268 #endif
269 	    UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
270 	if (nmbclusters > 0)
271 		uma_zone_set_max(zone_clust, nmbclusters);
272 
273 	zone_pack = uma_zsecond_create(MBUF_PACKET_MEM_NAME, mb_ctor_pack,
274 	    mb_dtor_pack, mb_zinit_pack, mb_zfini_pack, zone_mbuf);
275 
276 	/* Make jumbo frame zone too. Page size, 9k and 16k. */
277 	zone_jumbop = uma_zcreate(MBUF_JUMBOP_MEM_NAME, MJUMPAGESIZE,
278 	    mb_ctor_clust, mb_dtor_clust,
279 #ifdef INVARIANTS
280 	    trash_init, trash_fini,
281 #else
282 	    NULL, NULL,
283 #endif
284 	    UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
285 	if (nmbjumbop > 0)
286 		uma_zone_set_max(zone_jumbop, nmbjumbop);
287 
288 	zone_jumbo9 = uma_zcreate(MBUF_JUMBO9_MEM_NAME, MJUM9BYTES,
289 	    mb_ctor_clust, mb_dtor_clust,
290 #ifdef INVARIANTS
291 	    trash_init, trash_fini,
292 #else
293 	    NULL, NULL,
294 #endif
295 	    UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
296 	if (nmbjumbo9 > 0)
297 		uma_zone_set_max(zone_jumbo9, nmbjumbo9);
298 	uma_zone_set_allocf(zone_jumbo9, mbuf_jumbo_alloc);
299 	uma_zone_set_freef(zone_jumbo9, mbuf_jumbo_free);
300 
301 	zone_jumbo16 = uma_zcreate(MBUF_JUMBO16_MEM_NAME, MJUM16BYTES,
302 	    mb_ctor_clust, mb_dtor_clust,
303 #ifdef INVARIANTS
304 	    trash_init, trash_fini,
305 #else
306 	    NULL, NULL,
307 #endif
308 	    UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
309 	if (nmbjumbo16 > 0)
310 		uma_zone_set_max(zone_jumbo16, nmbjumbo16);
311 	uma_zone_set_allocf(zone_jumbo16, mbuf_jumbo_alloc);
312 	uma_zone_set_freef(zone_jumbo16, mbuf_jumbo_free);
313 
314 	zone_ext_refcnt = uma_zcreate(MBUF_EXTREFCNT_MEM_NAME, sizeof(u_int),
315 	    NULL, NULL,
316 	    NULL, NULL,
317 	    UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
318 
319 	/* uma_prealloc() goes here... */
320 
321 	/*
322 	 * Hook event handler for low-memory situation, used to
323 	 * drain protocols and push data back to the caches (UMA
324 	 * later pushes it back to VM).
325 	 */
326 	EVENTHANDLER_REGISTER(vm_lowmem, mb_reclaim, NULL,
327 	    EVENTHANDLER_PRI_FIRST);
328 
329 	/*
330 	 * [Re]set counters and local statistics knobs.
331 	 * XXX Some of these should go and be replaced, but UMA stat
332 	 * gathering needs to be revised.
333 	 */
334 	mbstat.m_mbufs = 0;
335 	mbstat.m_mclusts = 0;
336 	mbstat.m_drain = 0;
337 	mbstat.m_msize = MSIZE;
338 	mbstat.m_mclbytes = MCLBYTES;
339 	mbstat.m_minclsize = MINCLSIZE;
340 	mbstat.m_mlen = MLEN;
341 	mbstat.m_mhlen = MHLEN;
342 	mbstat.m_numtypes = MT_NTYPES;
343 
344 	mbstat.m_mcfail = mbstat.m_mpfail = 0;
345 	mbstat.sf_iocnt = 0;
346 	mbstat.sf_allocwait = mbstat.sf_allocfail = 0;
347 }
348 
349 /*
350  * UMA backend page allocator for the jumbo frame zones.
351  *
352  * Allocates kernel virtual memory that is backed by contiguous physical
353  * pages.
354  */
355 static void *
356 mbuf_jumbo_alloc(uma_zone_t zone, int bytes, u_int8_t *flags, int wait)
357 {
358 
359 	/* Inform UMA that this allocator uses kernel_map/object. */
360 	*flags = UMA_SLAB_KERNEL;
361 	return (contigmalloc(bytes, M_JUMBOFRAME, wait, (vm_paddr_t)0,
362 	    ~(vm_paddr_t)0, 1, 0));
363 }
364 
365 /*
366  * UMA backend page deallocator for the jumbo frame zones.
367  */
368 static void
369 mbuf_jumbo_free(void *mem, int size, u_int8_t flags)
370 {
371 
372 	contigfree(mem, size, M_JUMBOFRAME);
373 }
374 
375 /*
376  * Constructor for Mbuf master zone.
377  *
378  * The 'arg' pointer points to a mb_args structure which
379  * contains call-specific information required to support the
380  * mbuf allocation API.  See mbuf.h.
381  */
382 static int
383 mb_ctor_mbuf(void *mem, int size, void *arg, int how)
384 {
385 	struct mbuf *m;
386 	struct mb_args *args;
387 #ifdef MAC
388 	int error;
389 #endif
390 	int flags;
391 	short type;
392 
393 #ifdef INVARIANTS
394 	trash_ctor(mem, size, arg, how);
395 #endif
396 	m = (struct mbuf *)mem;
397 	args = (struct mb_args *)arg;
398 	flags = args->flags;
399 	type = args->type;
400 
401 	/*
402 	 * The mbuf is initialized later.  The caller has the
403 	 * responsibility to set up any MAC labels too.
404 	 */
405 	if (type == MT_NOINIT)
406 		return (0);
407 
408 	m->m_next = NULL;
409 	m->m_nextpkt = NULL;
410 	m->m_len = 0;
411 	m->m_flags = flags;
412 	m->m_type = type;
413 	if (flags & M_PKTHDR) {
414 		m->m_data = m->m_pktdat;
415 		m->m_pkthdr.rcvif = NULL;
416 		m->m_pkthdr.header = NULL;
417 		m->m_pkthdr.len = 0;
418 		m->m_pkthdr.csum_flags = 0;
419 		m->m_pkthdr.csum_data = 0;
420 		m->m_pkthdr.tso_segsz = 0;
421 		m->m_pkthdr.ether_vtag = 0;
422 		m->m_pkthdr.flowid = 0;
423 		SLIST_INIT(&m->m_pkthdr.tags);
424 #ifdef MAC
425 		/* If the label init fails, fail the alloc */
426 		error = mac_mbuf_init(m, how);
427 		if (error)
428 			return (error);
429 #endif
430 	} else
431 		m->m_data = m->m_dat;
432 	return (0);
433 }
434 
435 /*
436  * The Mbuf master zone destructor.
437  */
438 static void
439 mb_dtor_mbuf(void *mem, int size, void *arg)
440 {
441 	struct mbuf *m;
442 	unsigned long flags;
443 
444 	m = (struct mbuf *)mem;
445 	flags = (unsigned long)arg;
446 
447 	if ((flags & MB_NOTAGS) == 0 && (m->m_flags & M_PKTHDR) != 0)
448 		m_tag_delete_chain(m, NULL);
449 	KASSERT((m->m_flags & M_EXT) == 0, ("%s: M_EXT set", __func__));
450 	KASSERT((m->m_flags & M_NOFREE) == 0, ("%s: M_NOFREE set", __func__));
451 #ifdef INVARIANTS
452 	trash_dtor(mem, size, arg);
453 #endif
454 }
455 
456 /*
457  * The Mbuf Packet zone destructor.
458  */
459 static void
460 mb_dtor_pack(void *mem, int size, void *arg)
461 {
462 	struct mbuf *m;
463 
464 	m = (struct mbuf *)mem;
465 	if ((m->m_flags & M_PKTHDR) != 0)
466 		m_tag_delete_chain(m, NULL);
467 
468 	/* Make sure we've got a clean cluster back. */
469 	KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__));
470 	KASSERT(m->m_ext.ext_buf != NULL, ("%s: ext_buf == NULL", __func__));
471 	KASSERT(m->m_ext.ext_free == NULL, ("%s: ext_free != NULL", __func__));
472 	KASSERT(m->m_ext.ext_arg1 == NULL, ("%s: ext_arg1 != NULL", __func__));
473 	KASSERT(m->m_ext.ext_arg2 == NULL, ("%s: ext_arg2 != NULL", __func__));
474 	KASSERT(m->m_ext.ext_size == MCLBYTES, ("%s: ext_size != MCLBYTES", __func__));
475 	KASSERT(m->m_ext.ext_type == EXT_PACKET, ("%s: ext_type != EXT_PACKET", __func__));
476 	KASSERT(*m->m_ext.ref_cnt == 1, ("%s: ref_cnt != 1", __func__));
477 #ifdef INVARIANTS
478 	trash_dtor(m->m_ext.ext_buf, MCLBYTES, arg);
479 #endif
480 	/*
481 	 * If there are processes blocked on zone_clust, waiting for pages
482 	 * to be freed up, * cause them to be woken up by draining the
483 	 * packet zone.  We are exposed to a race here * (in the check for
484 	 * the UMA_ZFLAG_FULL) where we might miss the flag set, but that
485 	 * is deliberate. We don't want to acquire the zone lock for every
486 	 * mbuf free.
487 	 */
488 	if (uma_zone_exhausted_nolock(zone_clust))
489 		zone_drain(zone_pack);
490 }
491 
492 /*
493  * The Cluster and Jumbo[PAGESIZE|9|16] zone constructor.
494  *
495  * Here the 'arg' pointer points to the Mbuf which we
496  * are configuring cluster storage for.  If 'arg' is
497  * empty we allocate just the cluster without setting
498  * the mbuf to it.  See mbuf.h.
499  */
500 static int
501 mb_ctor_clust(void *mem, int size, void *arg, int how)
502 {
503 	struct mbuf *m;
504 	u_int *refcnt;
505 	int type;
506 	uma_zone_t zone;
507 
508 #ifdef INVARIANTS
509 	trash_ctor(mem, size, arg, how);
510 #endif
511 	switch (size) {
512 	case MCLBYTES:
513 		type = EXT_CLUSTER;
514 		zone = zone_clust;
515 		break;
516 #if MJUMPAGESIZE != MCLBYTES
517 	case MJUMPAGESIZE:
518 		type = EXT_JUMBOP;
519 		zone = zone_jumbop;
520 		break;
521 #endif
522 	case MJUM9BYTES:
523 		type = EXT_JUMBO9;
524 		zone = zone_jumbo9;
525 		break;
526 	case MJUM16BYTES:
527 		type = EXT_JUMBO16;
528 		zone = zone_jumbo16;
529 		break;
530 	default:
531 		panic("unknown cluster size");
532 		break;
533 	}
534 
535 	m = (struct mbuf *)arg;
536 	refcnt = uma_find_refcnt(zone, mem);
537 	*refcnt = 1;
538 	if (m != NULL) {
539 		m->m_ext.ext_buf = (caddr_t)mem;
540 		m->m_data = m->m_ext.ext_buf;
541 		m->m_flags |= M_EXT;
542 		m->m_ext.ext_free = NULL;
543 		m->m_ext.ext_arg1 = NULL;
544 		m->m_ext.ext_arg2 = NULL;
545 		m->m_ext.ext_size = size;
546 		m->m_ext.ext_type = type;
547 		m->m_ext.ref_cnt = refcnt;
548 	}
549 
550 	return (0);
551 }
552 
553 /*
554  * The Mbuf Cluster zone destructor.
555  */
556 static void
557 mb_dtor_clust(void *mem, int size, void *arg)
558 {
559 #ifdef INVARIANTS
560 	uma_zone_t zone;
561 
562 	zone = m_getzone(size);
563 	KASSERT(*(uma_find_refcnt(zone, mem)) <= 1,
564 		("%s: refcnt incorrect %u", __func__,
565 		 *(uma_find_refcnt(zone, mem))) );
566 
567 	trash_dtor(mem, size, arg);
568 #endif
569 }
570 
571 /*
572  * The Packet secondary zone's init routine, executed on the
573  * object's transition from mbuf keg slab to zone cache.
574  */
575 static int
576 mb_zinit_pack(void *mem, int size, int how)
577 {
578 	struct mbuf *m;
579 
580 	m = (struct mbuf *)mem;		/* m is virgin. */
581 	if (uma_zalloc_arg(zone_clust, m, how) == NULL ||
582 	    m->m_ext.ext_buf == NULL)
583 		return (ENOMEM);
584 	m->m_ext.ext_type = EXT_PACKET;	/* Override. */
585 #ifdef INVARIANTS
586 	trash_init(m->m_ext.ext_buf, MCLBYTES, how);
587 #endif
588 	return (0);
589 }
590 
591 /*
592  * The Packet secondary zone's fini routine, executed on the
593  * object's transition from zone cache to keg slab.
594  */
595 static void
596 mb_zfini_pack(void *mem, int size)
597 {
598 	struct mbuf *m;
599 
600 	m = (struct mbuf *)mem;
601 #ifdef INVARIANTS
602 	trash_fini(m->m_ext.ext_buf, MCLBYTES);
603 #endif
604 	uma_zfree_arg(zone_clust, m->m_ext.ext_buf, NULL);
605 #ifdef INVARIANTS
606 	trash_dtor(mem, size, NULL);
607 #endif
608 }
609 
610 /*
611  * The "packet" keg constructor.
612  */
613 static int
614 mb_ctor_pack(void *mem, int size, void *arg, int how)
615 {
616 	struct mbuf *m;
617 	struct mb_args *args;
618 #ifdef MAC
619 	int error;
620 #endif
621 	int flags;
622 	short type;
623 
624 	m = (struct mbuf *)mem;
625 	args = (struct mb_args *)arg;
626 	flags = args->flags;
627 	type = args->type;
628 
629 #ifdef INVARIANTS
630 	trash_ctor(m->m_ext.ext_buf, MCLBYTES, arg, how);
631 #endif
632 	m->m_next = NULL;
633 	m->m_nextpkt = NULL;
634 	m->m_data = m->m_ext.ext_buf;
635 	m->m_len = 0;
636 	m->m_flags = (flags | M_EXT);
637 	m->m_type = type;
638 
639 	if (flags & M_PKTHDR) {
640 		m->m_pkthdr.rcvif = NULL;
641 		m->m_pkthdr.len = 0;
642 		m->m_pkthdr.header = NULL;
643 		m->m_pkthdr.csum_flags = 0;
644 		m->m_pkthdr.csum_data = 0;
645 		m->m_pkthdr.tso_segsz = 0;
646 		m->m_pkthdr.ether_vtag = 0;
647 		m->m_pkthdr.flowid = 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