2 * SPDX-License-Identifier: BSD-2-Clause
4 * Copyright (c) 2004, 2005,
5 * Bosko Milekic <bmilekic@FreeBSD.org>. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice unmodified, this list of conditions and the following
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 #include <sys/cdefs.h>
31 #include "opt_param.h"
32 #include "opt_kern_tls.h"
34 #include <sys/param.h>
36 #include <sys/domainset.h>
37 #include <sys/malloc.h>
38 #include <sys/systm.h>
40 #include <sys/eventhandler.h>
41 #include <sys/kernel.h>
43 #include <sys/limits.h>
45 #include <sys/mutex.h>
46 #include <sys/refcount.h>
47 #include <sys/sf_buf.h>
49 #include <sys/socket.h>
50 #include <sys/sysctl.h>
53 #include <net/if_var.h>
56 #include <vm/vm_extern.h>
57 #include <vm/vm_kern.h>
58 #include <vm/vm_page.h>
59 #include <vm/vm_pageout.h>
60 #include <vm/vm_map.h>
62 #include <vm/uma_dbg.h>
64 _Static_assert(MJUMPAGESIZE > MCLBYTES,
65 "Cluster must be smaller than a jumbo page");
68 * In FreeBSD, Mbufs and Mbuf Clusters are allocated from UMA
71 * Mbuf Clusters (2K, contiguous) are allocated from the Cluster
72 * Zone. The Zone can be capped at kern.ipc.nmbclusters, if the
73 * administrator so desires.
75 * Mbufs are allocated from a UMA Primary Zone called the Mbuf
78 * Additionally, FreeBSD provides a Packet Zone, which it
79 * configures as a Secondary Zone to the Mbuf Primary Zone,
80 * thus sharing backend Slab kegs with the Mbuf Primary Zone.
82 * Thus common-case allocations and locking are simplified:
86 * | .------------>[(Packet Cache)] m_get(), m_gethdr()
88 * [(Cluster Cache)] [ Secondary ] [ (Mbuf Cache) ]
89 * [ Cluster Zone ] [ Zone ] [ Mbuf Primary Zone ]
95 * \____________(VM)_________________/
98 * Whenever an object is allocated with uma_zalloc() out of
99 * one of the Zones its _ctor_ function is executed. The same
100 * for any deallocation through uma_zfree() the _dtor_ function
103 * Caches are per-CPU and are filled from the Primary Zone.
105 * Whenever an object is allocated from the underlying global
106 * memory pool it gets pre-initialized with the _zinit_ functions.
107 * When the Keg's are overfull objects get decommissioned with
108 * _zfini_ functions and free'd back to the global memory pool.
112 int nmbufs; /* limits number of mbufs */
113 int nmbclusters; /* limits number of mbuf clusters */
114 int nmbjumbop; /* limits number of page size jumbo clusters */
115 int nmbjumbo9; /* limits number of 9k jumbo clusters */
116 int nmbjumbo16; /* limits number of 16k jumbo clusters */
118 bool mb_use_ext_pgs = false; /* use M_EXTPG mbufs for sendfile & TLS */
121 sysctl_mb_use_ext_pgs(SYSCTL_HANDLER_ARGS)
125 extpg = mb_use_ext_pgs;
126 error = sysctl_handle_int(oidp, &extpg, 0, req);
127 if (error == 0 && req->newptr != NULL) {
128 if (extpg != 0 && !PMAP_HAS_DMAP)
131 mb_use_ext_pgs = extpg != 0;
135 SYSCTL_PROC(_kern_ipc, OID_AUTO, mb_use_ext_pgs, CTLTYPE_INT | CTLFLAG_RW,
137 sysctl_mb_use_ext_pgs, "IU",
138 "Use unmapped mbufs for sendfile(2) and TLS offload");
140 static quad_t maxmbufmem; /* overall real memory limit for all mbufs */
142 SYSCTL_QUAD(_kern_ipc, OID_AUTO, maxmbufmem, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &maxmbufmem, 0,
143 "Maximum real memory allocatable to various mbuf types");
145 static counter_u64_t snd_tag_count;
146 SYSCTL_COUNTER_U64(_kern_ipc, OID_AUTO, num_snd_tags, CTLFLAG_RW,
147 &snd_tag_count, "# of active mbuf send tags");
150 * tunable_mbinit() has to be run before any mbuf allocations are done.
153 tunable_mbinit(void *dummy)
159 * The default limit for all mbuf related memory is 1/2 of all
160 * available kernel memory (physical or kmem).
161 * At most it can be 3/4 of available kernel memory.
163 realmem = qmin((quad_t)physmem * PAGE_SIZE, vm_kmem_size);
164 maxmbufmem = realmem / 2;
165 TUNABLE_QUAD_FETCH("kern.ipc.maxmbufmem", &maxmbufmem);
166 if (maxmbufmem > realmem / 4 * 3)
167 maxmbufmem = realmem / 4 * 3;
169 TUNABLE_INT_FETCH("kern.ipc.nmbclusters", &nmbclusters);
170 if (nmbclusters == 0)
171 nmbclusters = maxmbufmem / MCLBYTES / 4;
173 TUNABLE_INT_FETCH("kern.ipc.nmbjumbop", &nmbjumbop);
175 nmbjumbop = maxmbufmem / MJUMPAGESIZE / 4;
177 TUNABLE_INT_FETCH("kern.ipc.nmbjumbo9", &nmbjumbo9);
179 nmbjumbo9 = maxmbufmem / MJUM9BYTES / 6;
181 TUNABLE_INT_FETCH("kern.ipc.nmbjumbo16", &nmbjumbo16);
183 nmbjumbo16 = maxmbufmem / MJUM16BYTES / 6;
186 * We need at least as many mbufs as we have clusters of
187 * the various types added together.
189 TUNABLE_INT_FETCH("kern.ipc.nmbufs", &nmbufs);
190 if (nmbufs < nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16)
191 nmbufs = lmax(maxmbufmem / MSIZE / 5,
192 nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16);
195 * Unmapped mbufs can only safely be used on platforms with a direct
200 TUNABLE_INT_FETCH("kern.ipc.mb_use_ext_pgs", &extpg);
201 mb_use_ext_pgs = extpg != 0;
204 SYSINIT(tunable_mbinit, SI_SUB_KMEM, SI_ORDER_MIDDLE, tunable_mbinit, NULL);
207 sysctl_nmbclusters(SYSCTL_HANDLER_ARGS)
209 int error, newnmbclusters;
211 newnmbclusters = nmbclusters;
212 error = sysctl_handle_int(oidp, &newnmbclusters, 0, req);
213 if (error == 0 && req->newptr && newnmbclusters != nmbclusters) {
214 if (newnmbclusters > nmbclusters &&
215 nmbufs >= nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16) {
216 nmbclusters = newnmbclusters;
217 nmbclusters = uma_zone_set_max(zone_clust, nmbclusters);
218 EVENTHANDLER_INVOKE(nmbclusters_change);
224 SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbclusters,
225 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, &nmbclusters, 0,
226 sysctl_nmbclusters, "IU",
227 "Maximum number of mbuf clusters allowed");
230 sysctl_nmbjumbop(SYSCTL_HANDLER_ARGS)
232 int error, newnmbjumbop;
234 newnmbjumbop = nmbjumbop;
235 error = sysctl_handle_int(oidp, &newnmbjumbop, 0, req);
236 if (error == 0 && req->newptr && newnmbjumbop != nmbjumbop) {
237 if (newnmbjumbop > nmbjumbop &&
238 nmbufs >= nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16) {
239 nmbjumbop = newnmbjumbop;
240 nmbjumbop = uma_zone_set_max(zone_jumbop, nmbjumbop);
246 SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbop,
247 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, &nmbjumbop, 0,
248 sysctl_nmbjumbop, "IU",
249 "Maximum number of mbuf page size jumbo clusters allowed");
252 sysctl_nmbjumbo9(SYSCTL_HANDLER_ARGS)
254 int error, newnmbjumbo9;
256 newnmbjumbo9 = nmbjumbo9;
257 error = sysctl_handle_int(oidp, &newnmbjumbo9, 0, req);
258 if (error == 0 && req->newptr && newnmbjumbo9 != nmbjumbo9) {
259 if (newnmbjumbo9 > nmbjumbo9 &&
260 nmbufs >= nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16) {
261 nmbjumbo9 = newnmbjumbo9;
262 nmbjumbo9 = uma_zone_set_max(zone_jumbo9, nmbjumbo9);
268 SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbo9,
269 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, &nmbjumbo9, 0,
270 sysctl_nmbjumbo9, "IU",
271 "Maximum number of mbuf 9k jumbo clusters allowed");
274 sysctl_nmbjumbo16(SYSCTL_HANDLER_ARGS)
276 int error, newnmbjumbo16;
278 newnmbjumbo16 = nmbjumbo16;
279 error = sysctl_handle_int(oidp, &newnmbjumbo16, 0, req);
280 if (error == 0 && req->newptr && newnmbjumbo16 != nmbjumbo16) {
281 if (newnmbjumbo16 > nmbjumbo16 &&
282 nmbufs >= nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16) {
283 nmbjumbo16 = newnmbjumbo16;
284 nmbjumbo16 = uma_zone_set_max(zone_jumbo16, nmbjumbo16);
290 SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbo16,
291 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, &nmbjumbo16, 0,
292 sysctl_nmbjumbo16, "IU",
293 "Maximum number of mbuf 16k jumbo clusters allowed");
296 sysctl_nmbufs(SYSCTL_HANDLER_ARGS)
298 int error, newnmbufs;
301 error = sysctl_handle_int(oidp, &newnmbufs, 0, req);
302 if (error == 0 && req->newptr && newnmbufs != nmbufs) {
303 if (newnmbufs > nmbufs) {
305 nmbufs = uma_zone_set_max(zone_mbuf, nmbufs);
306 EVENTHANDLER_INVOKE(nmbufs_change);
312 SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbufs,
313 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
314 &nmbufs, 0, sysctl_nmbufs, "IU",
315 "Maximum number of mbufs allowed");
318 * Zones from which we allocate.
320 uma_zone_t zone_mbuf;
321 uma_zone_t zone_clust;
322 uma_zone_t zone_pack;
323 uma_zone_t zone_jumbop;
324 uma_zone_t zone_jumbo9;
325 uma_zone_t zone_jumbo16;
330 static int mb_ctor_mbuf(void *, int, void *, int);
331 static int mb_ctor_clust(void *, int, void *, int);
332 static int mb_ctor_pack(void *, int, void *, int);
333 static void mb_dtor_mbuf(void *, int, void *);
334 static void mb_dtor_pack(void *, int, void *);
335 static int mb_zinit_pack(void *, int, int);
336 static void mb_zfini_pack(void *, int);
337 static void mb_reclaim(uma_zone_t, int);
339 /* Ensure that MSIZE is a power of 2. */
340 CTASSERT((((MSIZE - 1) ^ MSIZE) + 1) >> 1 == MSIZE);
342 _Static_assert(sizeof(struct mbuf) <= MSIZE,
343 "size of mbuf exceeds MSIZE");
345 * Initialize FreeBSD Network buffer allocation.
348 mbuf_init(void *dummy)
352 * Configure UMA zones for Mbufs, Clusters, and Packets.
354 zone_mbuf = uma_zcreate(MBUF_MEM_NAME, MSIZE,
355 mb_ctor_mbuf, mb_dtor_mbuf, NULL, NULL,
356 MSIZE - 1, UMA_ZONE_CONTIG | UMA_ZONE_MAXBUCKET);
358 nmbufs = uma_zone_set_max(zone_mbuf, nmbufs);
359 uma_zone_set_warning(zone_mbuf, "kern.ipc.nmbufs limit reached");
360 uma_zone_set_maxaction(zone_mbuf, mb_reclaim);
362 zone_clust = uma_zcreate(MBUF_CLUSTER_MEM_NAME, MCLBYTES,
363 mb_ctor_clust, NULL, NULL, NULL,
364 UMA_ALIGN_PTR, UMA_ZONE_CONTIG);
366 nmbclusters = uma_zone_set_max(zone_clust, nmbclusters);
367 uma_zone_set_warning(zone_clust, "kern.ipc.nmbclusters limit reached");
368 uma_zone_set_maxaction(zone_clust, mb_reclaim);
370 zone_pack = uma_zsecond_create(MBUF_PACKET_MEM_NAME, mb_ctor_pack,
371 mb_dtor_pack, mb_zinit_pack, mb_zfini_pack, zone_mbuf);
373 /* Make jumbo frame zone too. Page size, 9k and 16k. */
374 zone_jumbop = uma_zcreate(MBUF_JUMBOP_MEM_NAME, MJUMPAGESIZE,
375 mb_ctor_clust, NULL, NULL, NULL,
376 UMA_ALIGN_PTR, UMA_ZONE_CONTIG);
378 nmbjumbop = uma_zone_set_max(zone_jumbop, nmbjumbop);
379 uma_zone_set_warning(zone_jumbop, "kern.ipc.nmbjumbop limit reached");
380 uma_zone_set_maxaction(zone_jumbop, mb_reclaim);
382 zone_jumbo9 = uma_zcreate(MBUF_JUMBO9_MEM_NAME, MJUM9BYTES,
383 mb_ctor_clust, NULL, NULL, NULL,
384 UMA_ALIGN_PTR, UMA_ZONE_CONTIG);
386 nmbjumbo9 = uma_zone_set_max(zone_jumbo9, nmbjumbo9);
387 uma_zone_set_warning(zone_jumbo9, "kern.ipc.nmbjumbo9 limit reached");
388 uma_zone_set_maxaction(zone_jumbo9, mb_reclaim);
390 zone_jumbo16 = uma_zcreate(MBUF_JUMBO16_MEM_NAME, MJUM16BYTES,
391 mb_ctor_clust, NULL, NULL, NULL,
392 UMA_ALIGN_PTR, UMA_ZONE_CONTIG);
394 nmbjumbo16 = uma_zone_set_max(zone_jumbo16, nmbjumbo16);
395 uma_zone_set_warning(zone_jumbo16, "kern.ipc.nmbjumbo16 limit reached");
396 uma_zone_set_maxaction(zone_jumbo16, mb_reclaim);
398 snd_tag_count = counter_u64_alloc(M_WAITOK);
400 SYSINIT(mbuf, SI_SUB_MBUF, SI_ORDER_FIRST, mbuf_init, NULL);
404 * debugnet makes use of a pre-allocated pool of mbufs and clusters. When
405 * debugnet is configured, we initialize a set of UMA cache zones which return
406 * items from this pool. At panic-time, the regular UMA zone pointers are
407 * overwritten with those of the cache zones so that drivers may allocate and
408 * free mbufs and clusters without attempting to allocate physical memory.
410 * We keep mbufs and clusters in a pair of mbuf queues. In particular, for
411 * the purpose of caching clusters, we treat them as mbufs.
413 static struct mbufq dn_mbufq =
414 { STAILQ_HEAD_INITIALIZER(dn_mbufq.mq_head), 0, INT_MAX };
415 static struct mbufq dn_clustq =
416 { STAILQ_HEAD_INITIALIZER(dn_clustq.mq_head), 0, INT_MAX };
418 static int dn_clsize;
419 static uma_zone_t dn_zone_mbuf;
420 static uma_zone_t dn_zone_clust;
421 static uma_zone_t dn_zone_pack;
423 static struct debugnet_saved_zones {
425 uma_zone_t dsz_clust;
427 uma_zone_t dsz_jumbop;
428 uma_zone_t dsz_jumbo9;
429 uma_zone_t dsz_jumbo16;
430 bool dsz_debugnet_zones_enabled;
434 dn_buf_import(void *arg, void **store, int count, int domain __unused,
443 for (i = 0; i < count; i++) {
444 m = mbufq_dequeue(q);
447 trash_init(m, q == &dn_mbufq ? MSIZE : dn_clsize, flags);
450 KASSERT((flags & M_WAITOK) == 0 || i == count,
451 ("%s: ran out of pre-allocated mbufs", __func__));
456 dn_buf_release(void *arg, void **store, int count)
464 for (i = 0; i < count; i++) {
466 (void)mbufq_enqueue(q, m);
471 dn_pack_import(void *arg __unused, void **store, int count, int domain __unused,
478 for (i = 0; i < count; i++) {
479 m = m_get(M_NOWAIT, MT_DATA);
482 clust = uma_zalloc(dn_zone_clust, M_NOWAIT);
487 mb_ctor_clust(clust, dn_clsize, m, 0);
490 KASSERT((flags & M_WAITOK) == 0 || i == count,
491 ("%s: ran out of pre-allocated mbufs", __func__));
496 dn_pack_release(void *arg __unused, void **store, int count)
502 for (i = 0; i < count; i++) {
504 clust = m->m_ext.ext_buf;
505 uma_zfree(dn_zone_clust, clust);
506 uma_zfree(dn_zone_mbuf, m);
511 * Free the pre-allocated mbufs and clusters reserved for debugnet, and destroy
512 * the corresponding UMA cache zones.
515 debugnet_mbuf_drain(void)
520 if (dn_zone_mbuf != NULL) {
521 uma_zdestroy(dn_zone_mbuf);
524 if (dn_zone_clust != NULL) {
525 uma_zdestroy(dn_zone_clust);
526 dn_zone_clust = NULL;
528 if (dn_zone_pack != NULL) {
529 uma_zdestroy(dn_zone_pack);
533 while ((m = mbufq_dequeue(&dn_mbufq)) != NULL)
535 while ((item = mbufq_dequeue(&dn_clustq)) != NULL)
536 uma_zfree(m_getzone(dn_clsize), item);
540 * Callback invoked immediately prior to starting a debugnet connection.
543 debugnet_mbuf_start(void)
546 MPASS(!dn_saved_zones.dsz_debugnet_zones_enabled);
548 /* Save the old zone pointers to restore when debugnet is closed. */
549 dn_saved_zones = (struct debugnet_saved_zones) {
550 .dsz_debugnet_zones_enabled = true,
551 .dsz_mbuf = zone_mbuf,
552 .dsz_clust = zone_clust,
553 .dsz_pack = zone_pack,
554 .dsz_jumbop = zone_jumbop,
555 .dsz_jumbo9 = zone_jumbo9,
556 .dsz_jumbo16 = zone_jumbo16,
560 * All cluster zones return buffers of the size requested by the
561 * drivers. It's up to the driver to reinitialize the zones if the
562 * MTU of a debugnet-enabled interface changes.
564 printf("debugnet: overwriting mbuf zone pointers\n");
565 zone_mbuf = dn_zone_mbuf;
566 zone_clust = dn_zone_clust;
567 zone_pack = dn_zone_pack;
568 zone_jumbop = dn_zone_clust;
569 zone_jumbo9 = dn_zone_clust;
570 zone_jumbo16 = dn_zone_clust;
574 * Callback invoked when a debugnet connection is closed/finished.
577 debugnet_mbuf_finish(void)
580 MPASS(dn_saved_zones.dsz_debugnet_zones_enabled);
582 printf("debugnet: restoring mbuf zone pointers\n");
583 zone_mbuf = dn_saved_zones.dsz_mbuf;
584 zone_clust = dn_saved_zones.dsz_clust;
585 zone_pack = dn_saved_zones.dsz_pack;
586 zone_jumbop = dn_saved_zones.dsz_jumbop;
587 zone_jumbo9 = dn_saved_zones.dsz_jumbo9;
588 zone_jumbo16 = dn_saved_zones.dsz_jumbo16;
590 memset(&dn_saved_zones, 0, sizeof(dn_saved_zones));
594 * Reinitialize the debugnet mbuf+cluster pool and cache zones.
597 debugnet_mbuf_reinit(int nmbuf, int nclust, int clsize)
602 debugnet_mbuf_drain();
606 dn_zone_mbuf = uma_zcache_create("debugnet_" MBUF_MEM_NAME,
607 MSIZE, mb_ctor_mbuf, mb_dtor_mbuf, NULL, NULL,
608 dn_buf_import, dn_buf_release,
609 &dn_mbufq, UMA_ZONE_NOBUCKET);
611 dn_zone_clust = uma_zcache_create("debugnet_" MBUF_CLUSTER_MEM_NAME,
612 clsize, mb_ctor_clust, NULL, NULL, NULL,
613 dn_buf_import, dn_buf_release,
614 &dn_clustq, UMA_ZONE_NOBUCKET);
616 dn_zone_pack = uma_zcache_create("debugnet_" MBUF_PACKET_MEM_NAME,
617 MCLBYTES, mb_ctor_pack, mb_dtor_pack, NULL, NULL,
618 dn_pack_import, dn_pack_release,
619 NULL, UMA_ZONE_NOBUCKET);
621 while (nmbuf-- > 0) {
622 m = m_get(M_WAITOK, MT_DATA);
623 uma_zfree(dn_zone_mbuf, m);
625 while (nclust-- > 0) {
626 item = uma_zalloc(m_getzone(dn_clsize), M_WAITOK);
627 uma_zfree(dn_zone_clust, item);
630 #endif /* DEBUGNET */
633 * Constructor for Mbuf primary zone.
635 * The 'arg' pointer points to a mb_args structure which
636 * contains call-specific information required to support the
637 * mbuf allocation API. See mbuf.h.
640 mb_ctor_mbuf(void *mem, int size, void *arg, int how)
643 struct mb_args *args;
648 args = (struct mb_args *)arg;
652 * The mbuf is initialized later. The caller has the
653 * responsibility to set up any MAC labels too.
655 if (type == MT_NOINIT)
658 m = (struct mbuf *)mem;
660 MPASS((flags & M_NOFREE) == 0);
662 error = m_init(m, how, type, flags);
668 * The Mbuf primary zone destructor.
671 mb_dtor_mbuf(void *mem, int size, void *arg)
674 unsigned long flags __diagused;
676 m = (struct mbuf *)mem;
677 flags = (unsigned long)arg;
679 KASSERT((m->m_flags & M_NOFREE) == 0, ("%s: M_NOFREE set", __func__));
680 KASSERT((flags & 0x1) == 0, ("%s: obsolete MB_DTOR_SKIP passed", __func__));
681 if ((m->m_flags & M_PKTHDR) && !SLIST_EMPTY(&m->m_pkthdr.tags))
682 m_tag_delete_chain(m, NULL);
686 * The Mbuf Packet zone destructor.
689 mb_dtor_pack(void *mem, int size, void *arg)
693 m = (struct mbuf *)mem;
694 if ((m->m_flags & M_PKTHDR) != 0)
695 m_tag_delete_chain(m, NULL);
697 /* Make sure we've got a clean cluster back. */
698 KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__));
699 KASSERT(m->m_ext.ext_buf != NULL, ("%s: ext_buf == NULL", __func__));
700 KASSERT(m->m_ext.ext_free == NULL, ("%s: ext_free != NULL", __func__));
701 KASSERT(m->m_ext.ext_arg1 == NULL, ("%s: ext_arg1 != NULL", __func__));
702 KASSERT(m->m_ext.ext_arg2 == NULL, ("%s: ext_arg2 != NULL", __func__));
703 KASSERT(m->m_ext.ext_size == MCLBYTES, ("%s: ext_size != MCLBYTES", __func__));
704 KASSERT(m->m_ext.ext_type == EXT_PACKET, ("%s: ext_type != EXT_PACKET", __func__));
705 #if defined(INVARIANTS) && !defined(KMSAN)
706 trash_dtor(m->m_ext.ext_buf, MCLBYTES, arg);
709 * If there are processes blocked on zone_clust, waiting for pages
710 * to be freed up, cause them to be woken up by draining the
711 * packet zone. We are exposed to a race here (in the check for
712 * the UMA_ZFLAG_FULL) where we might miss the flag set, but that
713 * is deliberate. We don't want to acquire the zone lock for every
716 if (uma_zone_exhausted(zone_clust))
717 uma_zone_reclaim(zone_pack, UMA_RECLAIM_DRAIN);
721 * The Cluster and Jumbo[PAGESIZE|9|16] zone constructor.
723 * Here the 'arg' pointer points to the Mbuf which we
724 * are configuring cluster storage for. If 'arg' is
725 * empty we allocate just the cluster without setting
726 * the mbuf to it. See mbuf.h.
729 mb_ctor_clust(void *mem, int size, void *arg, int how)
733 m = (struct mbuf *)arg;
735 m->m_ext.ext_buf = (char *)mem;
736 m->m_data = m->m_ext.ext_buf;
738 m->m_ext.ext_free = NULL;
739 m->m_ext.ext_arg1 = NULL;
740 m->m_ext.ext_arg2 = NULL;
741 m->m_ext.ext_size = size;
742 m->m_ext.ext_type = m_gettype(size);
743 m->m_ext.ext_flags = EXT_FLAG_EMBREF;
744 m->m_ext.ext_count = 1;
751 * The Packet secondary zone's init routine, executed on the
752 * object's transition from mbuf keg slab to zone cache.
755 mb_zinit_pack(void *mem, int size, int how)
759 m = (struct mbuf *)mem; /* m is virgin. */
760 if (uma_zalloc_arg(zone_clust, m, how) == NULL ||
761 m->m_ext.ext_buf == NULL)
763 m->m_ext.ext_type = EXT_PACKET; /* Override. */
764 #if defined(INVARIANTS) && !defined(KMSAN)
765 trash_init(m->m_ext.ext_buf, MCLBYTES, how);
771 * The Packet secondary zone's fini routine, executed on the
772 * object's transition from zone cache to keg slab.
775 mb_zfini_pack(void *mem, int size)
779 m = (struct mbuf *)mem;
780 #if defined(INVARIANTS) && !defined(KMSAN)
781 trash_fini(m->m_ext.ext_buf, MCLBYTES);
783 uma_zfree_arg(zone_clust, m->m_ext.ext_buf, NULL);
784 #if defined(INVARIANTS) && !defined(KMSAN)
785 trash_dtor(mem, size, NULL);
790 * The "packet" keg constructor.
793 mb_ctor_pack(void *mem, int size, void *arg, int how)
796 struct mb_args *args;
800 m = (struct mbuf *)mem;
801 args = (struct mb_args *)arg;
804 MPASS((flags & M_NOFREE) == 0);
806 #if defined(INVARIANTS) && !defined(KMSAN)
807 trash_ctor(m->m_ext.ext_buf, MCLBYTES, arg, how);
810 error = m_init(m, how, type, flags);
812 /* m_ext is already initialized. */
813 m->m_data = m->m_ext.ext_buf;
814 m->m_flags = (flags | M_EXT);
820 * This is the protocol drain routine. Called by UMA whenever any of the
821 * mbuf zones is closed to its limit.
824 mb_reclaim(uma_zone_t zone __unused, int pending __unused)
827 EVENTHANDLER_INVOKE(mbuf_lowmem, VM_LOW_MBUFS);
831 * Free "count" units of I/O from an mbuf chain. They could be held
832 * in M_EXTPG or just as a normal mbuf. This code is intended to be
833 * called in an error path (I/O error, closed connection, etc).
836 mb_free_notready(struct mbuf *m, int count)
840 for (i = 0; i < count && m != NULL; i++) {
841 if ((m->m_flags & M_EXTPG) != 0) {
843 if (m->m_epg_nrdy != 0)
848 KASSERT(i == count, ("Removed only %d items from %p", i, m));
852 * Compress an unmapped mbuf into a simple mbuf when it holds a small
853 * amount of data. This is used as a DOS defense to avoid having
854 * small packets tie up wired pages, an ext_pgs structure, and an
855 * mbuf. Since this converts the existing mbuf in place, it can only
856 * be used if there are no other references to 'm'.
859 mb_unmapped_compress(struct mbuf *m)
861 volatile u_int *refcnt;
865 * Assert that 'm' does not have a packet header. If 'm' had
866 * a packet header, it would only be able to hold MHLEN bytes
867 * and m_data would have to be initialized differently.
869 KASSERT((m->m_flags & M_PKTHDR) == 0 && (m->m_flags & M_EXTPG),
870 ("%s: m %p !M_EXTPG or M_PKTHDR", __func__, m));
871 KASSERT(m->m_len <= MLEN, ("m_len too large %p", m));
873 if (m->m_ext.ext_flags & EXT_FLAG_EMBREF) {
874 refcnt = &m->m_ext.ext_count;
876 KASSERT(m->m_ext.ext_cnt != NULL,
877 ("%s: no refcounting pointer on %p", __func__, m));
878 refcnt = m->m_ext.ext_cnt;
884 m_copydata(m, 0, m->m_len, buf);
886 /* Free the backing pages. */
887 m->m_ext.ext_free(m);
889 /* Turn 'm' into a "normal" mbuf. */
890 m->m_flags &= ~(M_EXT | M_RDONLY | M_EXTPG);
891 m->m_data = m->m_dat;
893 /* Copy data back into m. */
894 bcopy(buf, mtod(m, char *), m->m_len);
900 * These next few routines are used to permit downgrading an unmapped
901 * mbuf to a chain of mapped mbufs. This is used when an interface
902 * doesn't supported unmapped mbufs or if checksums need to be
903 * computed in software.
905 * Each unmapped mbuf is converted to a chain of mbufs. First, any
906 * TLS header data is stored in a regular mbuf. Second, each page of
907 * unmapped data is stored in an mbuf with an EXT_SFBUF external
908 * cluster. These mbufs use an sf_buf to provide a valid KVA for the
909 * associated physical page. They also hold a reference on the
910 * original M_EXTPG mbuf to ensure the physical page doesn't go away.
911 * Finally, any TLS trailer data is stored in a regular mbuf.
913 * mb_unmapped_free_mext() is the ext_free handler for the EXT_SFBUF
914 * mbufs. It frees the associated sf_buf and releases its reference
915 * on the original M_EXTPG mbuf.
917 * _mb_unmapped_to_ext() is a helper function that converts a single
918 * unmapped mbuf into a chain of mbufs.
920 * mb_unmapped_to_ext() is the public function that walks an mbuf
921 * chain converting any unmapped mbufs to mapped mbufs. It returns
922 * the new chain of unmapped mbufs on success. On failure it frees
923 * the original mbuf chain and returns NULL.
926 mb_unmapped_free_mext(struct mbuf *m)
931 sf = m->m_ext.ext_arg1;
934 /* Drop the reference on the backing M_EXTPG mbuf. */
935 old_m = m->m_ext.ext_arg2;
936 mb_free_extpg(old_m);
940 _mb_unmapped_to_ext(struct mbuf *m)
942 struct mbuf *m_new, *top, *prev, *mref;
945 int i, len, off, pglen, pgoff, seglen, segoff;
946 volatile u_int *refcnt;
951 KASSERT(m->m_epg_tls == NULL, ("%s: can't convert TLS mbuf %p",
954 /* See if this is the mbuf that holds the embedded refcount. */
955 if (m->m_ext.ext_flags & EXT_FLAG_EMBREF) {
956 refcnt = &m->m_ext.ext_count;
959 KASSERT(m->m_ext.ext_cnt != NULL,
960 ("%s: no refcounting pointer on %p", __func__, m));
961 refcnt = m->m_ext.ext_cnt;
962 mref = __containerof(refcnt, struct mbuf, m_ext.ext_count);
965 /* Skip over any data removed from the front. */
966 off = mtod(m, vm_offset_t);
969 if (m->m_epg_hdrlen != 0) {
970 if (off >= m->m_epg_hdrlen) {
971 off -= m->m_epg_hdrlen;
973 seglen = m->m_epg_hdrlen - off;
975 seglen = min(seglen, len);
978 m_new = m_get(M_NOWAIT, MT_DATA);
981 m_new->m_len = seglen;
983 memcpy(mtod(m_new, void *), &m->m_epg_hdr[segoff],
987 pgoff = m->m_epg_1st_off;
988 for (i = 0; i < m->m_epg_npgs && len > 0; i++) {
989 pglen = m_epg_pagelen(m, i, pgoff);
995 seglen = pglen - off;
996 segoff = pgoff + off;
998 seglen = min(seglen, len);
1001 pg = PHYS_TO_VM_PAGE(m->m_epg_pa[i]);
1002 m_new = m_get(M_NOWAIT, MT_DATA);
1008 prev->m_next = m_new;
1011 sf = sf_buf_alloc(pg, SFB_NOWAIT);
1016 m_extadd(m_new, (char *)sf_buf_kva(sf), PAGE_SIZE,
1017 mb_unmapped_free_mext, sf, mref, M_RDONLY, EXT_SFBUF);
1018 m_new->m_data += segoff;
1019 m_new->m_len = seglen;
1024 KASSERT((off + len) <= m->m_epg_trllen,
1025 ("off + len > trail (%d + %d > %d)", off, len,
1027 m_new = m_get(M_NOWAIT, MT_DATA);
1033 prev->m_next = m_new;
1035 memcpy(mtod(m_new, void *), &m->m_epg_trail[off], len);
1040 * Obtain an additional reference on the old mbuf for
1041 * each created EXT_SFBUF mbuf. They will be dropped
1042 * in mb_unmapped_free_mext().
1047 atomic_add_int(refcnt, ref_inc);
1055 * Obtain an additional reference on the old mbuf for
1056 * each created EXT_SFBUF mbuf. They will be
1057 * immediately dropped when these mbufs are freed
1063 atomic_add_int(refcnt, ref_inc);
1071 mb_unmapped_to_ext(struct mbuf *top)
1073 struct mbuf *m, *next, *prev = NULL;
1076 for (m = top; m != NULL; m = next) {
1077 /* m might be freed, so cache the next pointer. */
1079 if (m->m_flags & M_EXTPG) {
1082 * Remove 'm' from the new chain so
1083 * that the 'top' chain terminates
1084 * before 'm' in case 'top' is freed
1087 prev->m_next = NULL;
1089 m = _mb_unmapped_to_ext(m);
1102 * Replaced one mbuf with a chain, so we must
1103 * find the end of chain.
1117 * Allocate an empty M_EXTPG mbuf. The ext_free routine is
1118 * responsible for freeing any pages backing this mbuf when it is
1122 mb_alloc_ext_pgs(int how, m_ext_free_t ext_free)
1126 m = m_get(how, MT_DATA);
1132 m->m_epg_1st_off = 0;
1133 m->m_epg_last_len = 0;
1135 m->m_epg_hdrlen = 0;
1136 m->m_epg_trllen = 0;
1137 m->m_epg_tls = NULL;
1140 m->m_flags |= (M_EXT | M_RDONLY | M_EXTPG);
1141 m->m_ext.ext_flags = EXT_FLAG_EMBREF;
1142 m->m_ext.ext_count = 1;
1143 m->m_ext.ext_size = 0;
1144 m->m_ext.ext_free = ext_free;
1149 * Clean up after mbufs with M_EXT storage attached to them if the
1150 * reference count hits 1.
1153 mb_free_ext(struct mbuf *m)
1155 volatile u_int *refcnt;
1159 KASSERT(m->m_flags & M_EXT, ("%s: M_EXT not set on %p", __func__, m));
1161 /* See if this is the mbuf that holds the embedded refcount. */
1162 if (m->m_ext.ext_flags & EXT_FLAG_EMBREF) {
1163 refcnt = &m->m_ext.ext_count;
1166 KASSERT(m->m_ext.ext_cnt != NULL,
1167 ("%s: no refcounting pointer on %p", __func__, m));
1168 refcnt = m->m_ext.ext_cnt;
1169 mref = __containerof(refcnt, struct mbuf, m_ext.ext_count);
1173 * Check if the header is embedded in the cluster. It is
1174 * important that we can't touch any of the mbuf fields
1175 * after we have freed the external storage, since mbuf
1176 * could have been embedded in it. For now, the mbufs
1177 * embedded into the cluster are always of type EXT_EXTREF,
1178 * and for this type we won't free the mref.
1180 if (m->m_flags & M_NOFREE) {
1182 KASSERT(m->m_ext.ext_type == EXT_EXTREF ||
1183 m->m_ext.ext_type == EXT_RXRING,
1184 ("%s: no-free mbuf %p has wrong type", __func__, m));
1188 /* Free attached storage if this mbuf is the only reference to it. */
1189 if (*refcnt == 1 || atomic_fetchadd_int(refcnt, -1) == 1) {
1190 switch (m->m_ext.ext_type) {
1192 /* The packet zone is special. */
1195 uma_zfree(zone_pack, mref);
1198 uma_zfree(zone_clust, m->m_ext.ext_buf);
1202 uma_zfree(zone_jumbop, m->m_ext.ext_buf);
1206 uma_zfree(zone_jumbo9, m->m_ext.ext_buf);
1210 uma_zfree(zone_jumbo16, m->m_ext.ext_buf);
1216 case EXT_DISPOSABLE:
1217 KASSERT(mref->m_ext.ext_free != NULL,
1218 ("%s: ext_free not set", __func__));
1219 mref->m_ext.ext_free(mref);
1223 KASSERT(m->m_ext.ext_free != NULL,
1224 ("%s: ext_free not set", __func__));
1225 m->m_ext.ext_free(m);
1228 KASSERT(m->m_ext.ext_free == NULL,
1229 ("%s: ext_free is set", __func__));
1232 KASSERT(m->m_ext.ext_type == 0,
1233 ("%s: unknown ext_type", __func__));
1237 if (freembuf && m != mref)
1242 * Clean up after mbufs with M_EXTPG storage attached to them if the
1243 * reference count hits 1.
1246 mb_free_extpg(struct mbuf *m)
1248 volatile u_int *refcnt;
1253 /* See if this is the mbuf that holds the embedded refcount. */
1254 if (m->m_ext.ext_flags & EXT_FLAG_EMBREF) {
1255 refcnt = &m->m_ext.ext_count;
1258 KASSERT(m->m_ext.ext_cnt != NULL,
1259 ("%s: no refcounting pointer on %p", __func__, m));
1260 refcnt = m->m_ext.ext_cnt;
1261 mref = __containerof(refcnt, struct mbuf, m_ext.ext_count);
1264 /* Free attached storage if this mbuf is the only reference to it. */
1265 if (*refcnt == 1 || atomic_fetchadd_int(refcnt, -1) == 1) {
1266 KASSERT(mref->m_ext.ext_free != NULL,
1267 ("%s: ext_free not set", __func__));
1269 mref->m_ext.ext_free(mref);
1271 if (mref->m_epg_tls != NULL &&
1272 !refcount_release_if_not_last(&mref->m_epg_tls->refcount))
1273 ktls_enqueue_to_free(mref);
1284 * Official mbuf(9) allocation KPI for stack and drivers:
1286 * m_get() - a single mbuf without any attachments, sys/mbuf.h.
1287 * m_gethdr() - a single mbuf initialized as M_PKTHDR, sys/mbuf.h.
1288 * m_getcl() - an mbuf + 2k cluster, sys/mbuf.h.
1289 * m_clget() - attach cluster to already allocated mbuf.
1290 * m_cljget() - attach jumbo cluster to already allocated mbuf.
1291 * m_get2() - allocate minimum mbuf that would fit size argument.
1292 * m_getm2() - allocate a chain of mbufs/clusters.
1293 * m_extadd() - attach external cluster to mbuf.
1295 * m_free() - free single mbuf with its tags and ext, sys/mbuf.h.
1296 * m_freem() - free chain of mbufs.
1300 m_clget(struct mbuf *m, int how)
1303 KASSERT((m->m_flags & M_EXT) == 0, ("%s: mbuf %p has M_EXT",
1305 m->m_ext.ext_buf = (char *)NULL;
1306 uma_zalloc_arg(zone_clust, m, how);
1308 * On a cluster allocation failure, drain the packet zone and retry,
1309 * we might be able to loosen a few clusters up on the drain.
1311 if ((how & M_NOWAIT) && (m->m_ext.ext_buf == NULL)) {
1312 uma_zone_reclaim(zone_pack, UMA_RECLAIM_DRAIN);
1313 uma_zalloc_arg(zone_clust, m, how);
1315 MBUF_PROBE2(m__clget, m, how);
1316 return (m->m_flags & M_EXT);
1320 * m_cljget() is different from m_clget() as it can allocate clusters without
1321 * attaching them to an mbuf. In that case the return value is the pointer
1322 * to the cluster of the requested size. If an mbuf was specified, it gets
1323 * the cluster attached to it and the return value can be safely ignored.
1324 * For size it takes MCLBYTES, MJUMPAGESIZE, MJUM9BYTES, MJUM16BYTES.
1327 m_cljget(struct mbuf *m, int how, int size)
1333 KASSERT((m->m_flags & M_EXT) == 0, ("%s: mbuf %p has M_EXT",
1335 m->m_ext.ext_buf = NULL;
1338 zone = m_getzone(size);
1339 retval = uma_zalloc_arg(zone, m, how);
1341 MBUF_PROBE4(m__cljget, m, how, size, retval);
1347 * m_get2() allocates minimum mbuf that would fit "size" argument.
1350 m_get2(int size, int how, short type, int flags)
1352 struct mb_args args;
1358 if (size <= MHLEN || (size <= MLEN && (flags & M_PKTHDR) == 0))
1359 return (uma_zalloc_arg(zone_mbuf, &args, how));
1360 if (size <= MCLBYTES)
1361 return (uma_zalloc_arg(zone_pack, &args, how));
1363 if (size > MJUMPAGESIZE)
1366 m = uma_zalloc_arg(zone_mbuf, &args, how);
1370 n = uma_zalloc_arg(zone_jumbop, m, how);
1380 * m_get3() allocates minimum mbuf that would fit "size" argument.
1381 * Unlike m_get2() it can allocate clusters up to MJUM16BYTES.
1384 m_get3(int size, int how, short type, int flags)
1386 struct mb_args args;
1390 if (size <= MJUMPAGESIZE)
1391 return (m_get2(size, how, type, flags));
1393 if (size > MJUM16BYTES)
1399 m = uma_zalloc_arg(zone_mbuf, &args, how);
1403 if (size <= MJUM9BYTES)
1406 zone = zone_jumbo16;
1408 n = uma_zalloc_arg(zone, m, how);
1418 * m_getjcl() returns an mbuf with a cluster of the specified size attached.
1419 * For size it takes MCLBYTES, MJUMPAGESIZE, MJUM9BYTES, MJUM16BYTES.
1422 m_getjcl(int how, short type, int flags, int size)
1424 struct mb_args args;
1428 if (size == MCLBYTES)
1429 return m_getcl(how, type, flags);
1434 m = uma_zalloc_arg(zone_mbuf, &args, how);
1438 zone = m_getzone(size);
1439 n = uma_zalloc_arg(zone, m, how);
1444 MBUF_PROBE5(m__getjcl, how, type, flags, size, m);
1449 * Allocate a given length worth of mbufs and/or clusters (whatever fits
1450 * best) and return a pointer to the top of the allocated chain. If an
1451 * existing mbuf chain is provided, then we will append the new chain
1452 * to the existing one and return a pointer to the provided mbuf.
1455 m_getm2(struct mbuf *m, int len, int how, short type, int flags)
1457 struct mbuf *mb, *nm = NULL, *mtail = NULL;
1459 KASSERT(len >= 0, ("%s: len is < 0", __func__));
1461 /* Validate flags. */
1462 flags &= (M_PKTHDR | M_EOR);
1464 /* Packet header mbuf must be first in chain. */
1465 if ((flags & M_PKTHDR) && m != NULL)
1468 /* Loop and append maximum sized mbufs to the chain tail. */
1471 if (len > MCLBYTES) {
1472 mb = m_getjcl(M_NOWAIT, type, (flags & M_PKTHDR),
1476 if (len >= MINCLSIZE)
1477 mb = m_getcl(how, type, (flags & M_PKTHDR));
1478 else if (flags & M_PKTHDR)
1479 mb = m_gethdr(how, type);
1481 mb = m_get(how, type);
1484 * Fail the whole operation if one mbuf can't be
1500 flags &= ~M_PKTHDR; /* Only valid on the first mbuf. */
1503 mtail->m_flags |= M_EOR; /* Only valid on the last mbuf. */
1505 /* If mbuf was supplied, append new chain to the end of it. */
1507 for (mtail = m; mtail->m_next != NULL; mtail = mtail->m_next)
1510 mtail->m_flags &= ~M_EOR;
1518 * Configure a provided mbuf to refer to the provided external storage
1519 * buffer and setup a reference count for said buffer.
1522 * mb The existing mbuf to which to attach the provided buffer.
1523 * buf The address of the provided external storage buffer.
1524 * size The size of the provided buffer.
1525 * freef A pointer to a routine that is responsible for freeing the
1526 * provided external storage buffer.
1527 * args A pointer to an argument structure (of any type) to be passed
1528 * to the provided freef routine (may be NULL).
1529 * flags Any other flags to be passed to the provided mbuf.
1530 * type The type that the external storage buffer should be
1537 m_extadd(struct mbuf *mb, char *buf, u_int size, m_ext_free_t freef,
1538 void *arg1, void *arg2, int flags, int type)
1541 KASSERT(type != EXT_CLUSTER, ("%s: EXT_CLUSTER not allowed", __func__));
1543 mb->m_flags |= (M_EXT | flags);
1544 mb->m_ext.ext_buf = buf;
1545 mb->m_data = mb->m_ext.ext_buf;
1546 mb->m_ext.ext_size = size;
1547 mb->m_ext.ext_free = freef;
1548 mb->m_ext.ext_arg1 = arg1;
1549 mb->m_ext.ext_arg2 = arg2;
1550 mb->m_ext.ext_type = type;
1552 if (type != EXT_EXTREF) {
1553 mb->m_ext.ext_count = 1;
1554 mb->m_ext.ext_flags = EXT_FLAG_EMBREF;
1556 mb->m_ext.ext_flags = 0;
1560 * Free an entire chain of mbufs and associated external buffers, if
1564 m_freem(struct mbuf *mb)
1567 MBUF_PROBE1(m__freem, mb);
1573 * Temporary primitive to allow freeing without going through m_free.
1576 m_free_raw(struct mbuf *mb)
1579 uma_zfree(zone_mbuf, mb);
1583 m_snd_tag_alloc(struct ifnet *ifp, union if_snd_tag_alloc_params *params,
1584 struct m_snd_tag **mstp)
1587 return (if_snd_tag_alloc(ifp, params, mstp));
1591 m_snd_tag_init(struct m_snd_tag *mst, struct ifnet *ifp,
1592 const struct if_snd_tag_sw *sw)
1597 refcount_init(&mst->refcount, 1);
1599 counter_u64_add(snd_tag_count, 1);
1603 m_snd_tag_destroy(struct m_snd_tag *mst)
1608 mst->sw->snd_tag_free(mst);
1610 counter_u64_add(snd_tag_count, -1);
1614 m_rcvif_serialize(struct mbuf *m)
1619 idx = if_getindex(m->m_pkthdr.rcvif);
1620 gen = if_getidxgen(m->m_pkthdr.rcvif);
1621 m->m_pkthdr.rcvidx = idx;
1622 m->m_pkthdr.rcvgen = gen;
1623 if (__predict_false(m->m_pkthdr.leaf_rcvif != NULL)) {
1624 idx = if_getindex(m->m_pkthdr.leaf_rcvif);
1625 gen = if_getidxgen(m->m_pkthdr.leaf_rcvif);
1630 m->m_pkthdr.leaf_rcvidx = idx;
1631 m->m_pkthdr.leaf_rcvgen = gen;
1635 m_rcvif_restore(struct mbuf *m)
1637 struct ifnet *ifp, *leaf_ifp;
1642 ifp = ifnet_byindexgen(m->m_pkthdr.rcvidx, m->m_pkthdr.rcvgen);
1643 if (ifp == NULL || (if_getflags(ifp) & IFF_DYING))
1646 if (__predict_true(m->m_pkthdr.leaf_rcvidx == (u_short)-1)) {
1649 leaf_ifp = ifnet_byindexgen(m->m_pkthdr.leaf_rcvidx,
1650 m->m_pkthdr.leaf_rcvgen);
1651 if (__predict_false(leaf_ifp != NULL && (if_getflags(leaf_ifp) & IFF_DYING)))
1655 m->m_pkthdr.leaf_rcvif = leaf_ifp;
1656 m->m_pkthdr.rcvif = ifp;
1662 * Allocate an mbuf with anonymous external pages.
1665 mb_alloc_ext_plus_pages(int len, int how)
1671 m = mb_alloc_ext_pgs(how, mb_free_mext_pgs);
1674 m->m_epg_flags |= EPG_FLAG_ANON;
1675 npgs = howmany(len, PAGE_SIZE);
1676 for (i = 0; i < npgs; i++) {
1678 pg = vm_page_alloc_noobj(VM_ALLOC_NODUMP |
1681 if (how == M_NOWAIT) {
1688 } while (pg == NULL);
1689 m->m_epg_pa[i] = VM_PAGE_TO_PHYS(pg);
1691 m->m_epg_npgs = npgs;
1696 * Copy the data in the mbuf chain to a chain of mbufs with anonymous external
1698 * len is the length of data in the input mbuf chain.
1699 * mlen is the maximum number of bytes put into each ext_page mbuf.
1702 mb_mapped_to_unmapped(struct mbuf *mp, int len, int mlen, int how,
1703 struct mbuf **mlast)
1705 struct mbuf *m, *mout;
1706 char *pgpos, *mbpos;
1707 int i, mblen, mbufsiz, pglen, xfer;
1711 mbufsiz = min(mlen, len);
1712 m = mout = mb_alloc_ext_plus_pages(mbufsiz, how);
1715 pgpos = (char *)(void *)PHYS_TO_DMAP(m->m_epg_pa[0]);
1721 if (++i == m->m_epg_npgs) {
1722 m->m_epg_last_len = PAGE_SIZE;
1723 mbufsiz = min(mlen, len);
1724 m->m_next = mb_alloc_ext_plus_pages(mbufsiz,
1733 pgpos = (char *)(void *)PHYS_TO_DMAP(m->m_epg_pa[i]);
1736 while (mblen == 0) {
1741 KASSERT((mp->m_flags & M_EXTPG) == 0,
1742 ("mb_copym_ext_pgs: ext_pgs input mbuf"));
1743 mbpos = mtod(mp, char *);
1747 xfer = min(mblen, pglen);
1748 memcpy(pgpos, mbpos, xfer);
1756 m->m_epg_last_len = PAGE_SIZE - pglen;