2 * Copyright (c) 2004, 2005,
3 * Bosko Milekic <bmilekic@FreeBSD.org>. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice unmodified, this list of conditions and the following
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.
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
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
31 #include "opt_param.h"
33 #include <sys/param.h>
34 #include <sys/malloc.h>
35 #include <sys/systm.h>
37 #include <sys/domain.h>
38 #include <sys/eventhandler.h>
39 #include <sys/kernel.h>
40 #include <sys/protosw.h>
42 #include <sys/sysctl.h>
44 #include <security/mac/mac_framework.h>
47 #include <vm/vm_extern.h>
48 #include <vm/vm_kern.h>
49 #include <vm/vm_page.h>
50 #include <vm/vm_map.h>
52 #include <vm/uma_int.h>
53 #include <vm/uma_dbg.h>
56 * In FreeBSD, Mbufs and Mbuf Clusters are allocated from UMA
59 * Mbuf Clusters (2K, contiguous) are allocated from the Cluster
60 * Zone. The Zone can be capped at kern.ipc.nmbclusters, if the
61 * administrator so desires.
63 * Mbufs are allocated from a UMA Master Zone called the Mbuf
66 * Additionally, FreeBSD provides a Packet Zone, which it
67 * configures as a Secondary Zone to the Mbuf Master Zone,
68 * thus sharing backend Slab kegs with the Mbuf Master Zone.
70 * Thus common-case allocations and locking are simplified:
74 * | .------------>[(Packet Cache)] m_get(), m_gethdr()
76 * [(Cluster Cache)] [ Secondary ] [ (Mbuf Cache) ]
77 * [ Cluster Zone ] [ Zone ] [ Mbuf Master Zone ]
83 * \____________(VM)_________________/
86 * Whenever an object is allocated with uma_zalloc() out of
87 * one of the Zones its _ctor_ function is executed. The same
88 * for any deallocation through uma_zfree() the _dtor_ function
91 * Caches are per-CPU and are filled from the Master Zone.
93 * Whenever an object is allocated from the underlying global
94 * memory pool it gets pre-initialized with the _zinit_ functions.
95 * When the Keg's are overfull objects get decomissioned with
96 * _zfini_ functions and free'd back to the global memory pool.
100 int nmbufs; /* limits number of mbufs */
101 int nmbclusters; /* limits number of mbuf clusters */
102 int nmbjumbop; /* limits number of page size jumbo clusters */
103 int nmbjumbo9; /* limits number of 9k jumbo clusters */
104 int nmbjumbo16; /* limits number of 16k jumbo clusters */
106 static quad_t maxmbufmem; /* overall real memory limit for all mbufs */
108 SYSCTL_QUAD(_kern_ipc, OID_AUTO, maxmbufmem, CTLFLAG_RDTUN, &maxmbufmem, 0,
109 "Maximum real memory allocatable to various mbuf types");
112 * tunable_mbinit() has to be run before any mbuf allocations are done.
115 tunable_mbinit(void *dummy)
120 * The default limit for all mbuf related memory is 1/2 of all
121 * available kernel memory (physical or kmem).
122 * At most it can be 3/4 of available kernel memory.
124 realmem = qmin((quad_t)physmem * PAGE_SIZE, vm_kmem_size);
125 maxmbufmem = realmem / 2;
126 TUNABLE_QUAD_FETCH("kern.ipc.maxmbufmem", &maxmbufmem);
127 if (maxmbufmem > realmem / 4 * 3)
128 maxmbufmem = realmem / 4 * 3;
130 TUNABLE_INT_FETCH("kern.ipc.nmbclusters", &nmbclusters);
131 if (nmbclusters == 0)
132 nmbclusters = maxmbufmem / MCLBYTES / 4;
134 TUNABLE_INT_FETCH("kern.ipc.nmbjumbop", &nmbjumbop);
136 nmbjumbop = maxmbufmem / MJUMPAGESIZE / 4;
138 TUNABLE_INT_FETCH("kern.ipc.nmbjumbo9", &nmbjumbo9);
140 nmbjumbo9 = maxmbufmem / MJUM9BYTES / 6;
142 TUNABLE_INT_FETCH("kern.ipc.nmbjumbo16", &nmbjumbo16);
144 nmbjumbo16 = maxmbufmem / MJUM16BYTES / 6;
147 * We need at least as many mbufs as we have clusters of
148 * the various types added together.
150 TUNABLE_INT_FETCH("kern.ipc.nmbufs", &nmbufs);
151 if (nmbufs < nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16)
152 nmbufs = lmax(maxmbufmem / MSIZE / 5,
153 nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16);
155 SYSINIT(tunable_mbinit, SI_SUB_KMEM, SI_ORDER_MIDDLE, tunable_mbinit, NULL);
158 sysctl_nmbclusters(SYSCTL_HANDLER_ARGS)
160 int error, newnmbclusters;
162 newnmbclusters = nmbclusters;
163 error = sysctl_handle_int(oidp, &newnmbclusters, 0, req);
164 if (error == 0 && req->newptr && newnmbclusters != nmbclusters) {
165 if (newnmbclusters > nmbclusters &&
166 nmbufs >= nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16) {
167 nmbclusters = newnmbclusters;
168 nmbclusters = uma_zone_set_max(zone_clust, nmbclusters);
169 EVENTHANDLER_INVOKE(nmbclusters_change);
175 SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbclusters, CTLTYPE_INT|CTLFLAG_RW,
176 &nmbclusters, 0, sysctl_nmbclusters, "IU",
177 "Maximum number of mbuf clusters allowed");
180 sysctl_nmbjumbop(SYSCTL_HANDLER_ARGS)
182 int error, newnmbjumbop;
184 newnmbjumbop = nmbjumbop;
185 error = sysctl_handle_int(oidp, &newnmbjumbop, 0, req);
186 if (error == 0 && req->newptr && newnmbjumbop != nmbjumbop) {
187 if (newnmbjumbop > nmbjumbop &&
188 nmbufs >= nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16) {
189 nmbjumbop = newnmbjumbop;
190 nmbjumbop = uma_zone_set_max(zone_jumbop, nmbjumbop);
196 SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbop, CTLTYPE_INT|CTLFLAG_RW,
197 &nmbjumbop, 0, sysctl_nmbjumbop, "IU",
198 "Maximum number of mbuf page size jumbo clusters allowed");
201 sysctl_nmbjumbo9(SYSCTL_HANDLER_ARGS)
203 int error, newnmbjumbo9;
205 newnmbjumbo9 = nmbjumbo9;
206 error = sysctl_handle_int(oidp, &newnmbjumbo9, 0, req);
207 if (error == 0 && req->newptr && newnmbjumbo9 != nmbjumbo9) {
208 if (newnmbjumbo9 > nmbjumbo9 &&
209 nmbufs >= nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16) {
210 nmbjumbo9 = newnmbjumbo9;
211 nmbjumbo9 = uma_zone_set_max(zone_jumbo9, nmbjumbo9);
217 SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbo9, CTLTYPE_INT|CTLFLAG_RW,
218 &nmbjumbo9, 0, sysctl_nmbjumbo9, "IU",
219 "Maximum number of mbuf 9k jumbo clusters allowed");
222 sysctl_nmbjumbo16(SYSCTL_HANDLER_ARGS)
224 int error, newnmbjumbo16;
226 newnmbjumbo16 = nmbjumbo16;
227 error = sysctl_handle_int(oidp, &newnmbjumbo16, 0, req);
228 if (error == 0 && req->newptr && newnmbjumbo16 != nmbjumbo16) {
229 if (newnmbjumbo16 > nmbjumbo16 &&
230 nmbufs >= nmbclusters + nmbjumbop + nmbjumbo9 + nmbjumbo16) {
231 nmbjumbo16 = newnmbjumbo16;
232 nmbjumbo16 = uma_zone_set_max(zone_jumbo16, nmbjumbo16);
238 SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbjumbo16, CTLTYPE_INT|CTLFLAG_RW,
239 &nmbjumbo16, 0, sysctl_nmbjumbo16, "IU",
240 "Maximum number of mbuf 16k jumbo clusters allowed");
243 sysctl_nmbufs(SYSCTL_HANDLER_ARGS)
245 int error, newnmbufs;
248 error = sysctl_handle_int(oidp, &newnmbufs, 0, req);
249 if (error == 0 && req->newptr && newnmbufs != nmbufs) {
250 if (newnmbufs > nmbufs) {
252 nmbufs = uma_zone_set_max(zone_mbuf, nmbufs);
253 EVENTHANDLER_INVOKE(nmbufs_change);
259 SYSCTL_PROC(_kern_ipc, OID_AUTO, nmbufs, CTLTYPE_INT|CTLFLAG_RW,
260 &nmbufs, 0, sysctl_nmbufs, "IU",
261 "Maximum number of mbufs allowed");
264 * Zones from which we allocate.
266 uma_zone_t zone_mbuf;
267 uma_zone_t zone_clust;
268 uma_zone_t zone_pack;
269 uma_zone_t zone_jumbop;
270 uma_zone_t zone_jumbo9;
271 uma_zone_t zone_jumbo16;
272 uma_zone_t zone_ext_refcnt;
277 static int mb_ctor_mbuf(void *, int, void *, int);
278 static int mb_ctor_clust(void *, int, void *, int);
279 static int mb_ctor_pack(void *, int, void *, int);
280 static void mb_dtor_mbuf(void *, int, void *);
281 static void mb_dtor_clust(void *, int, void *);
282 static void mb_dtor_pack(void *, int, void *);
283 static int mb_zinit_pack(void *, int, int);
284 static void mb_zfini_pack(void *, int);
286 static void mb_reclaim(void *);
287 static void *mbuf_jumbo_alloc(uma_zone_t, int, uint8_t *, int);
289 /* Ensure that MSIZE is a power of 2. */
290 CTASSERT((((MSIZE - 1) ^ MSIZE) + 1) >> 1 == MSIZE);
293 * Initialize FreeBSD Network buffer allocation.
296 mbuf_init(void *dummy)
300 * Configure UMA zones for Mbufs, Clusters, and Packets.
302 zone_mbuf = uma_zcreate(MBUF_MEM_NAME, MSIZE,
303 mb_ctor_mbuf, mb_dtor_mbuf,
305 trash_init, trash_fini,
309 MSIZE - 1, UMA_ZONE_MAXBUCKET);
311 nmbufs = uma_zone_set_max(zone_mbuf, nmbufs);
312 uma_zone_set_warning(zone_mbuf, "kern.ipc.nmbufs limit reached");
314 zone_clust = uma_zcreate(MBUF_CLUSTER_MEM_NAME, MCLBYTES,
315 mb_ctor_clust, mb_dtor_clust,
317 trash_init, trash_fini,
321 UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
323 nmbclusters = uma_zone_set_max(zone_clust, nmbclusters);
324 uma_zone_set_warning(zone_clust, "kern.ipc.nmbclusters limit reached");
326 zone_pack = uma_zsecond_create(MBUF_PACKET_MEM_NAME, mb_ctor_pack,
327 mb_dtor_pack, mb_zinit_pack, mb_zfini_pack, zone_mbuf);
329 /* Make jumbo frame zone too. Page size, 9k and 16k. */
330 zone_jumbop = uma_zcreate(MBUF_JUMBOP_MEM_NAME, MJUMPAGESIZE,
331 mb_ctor_clust, mb_dtor_clust,
333 trash_init, trash_fini,
337 UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
339 nmbjumbop = uma_zone_set_max(zone_jumbop, nmbjumbop);
340 uma_zone_set_warning(zone_jumbop, "kern.ipc.nmbjumbop limit reached");
342 zone_jumbo9 = uma_zcreate(MBUF_JUMBO9_MEM_NAME, MJUM9BYTES,
343 mb_ctor_clust, mb_dtor_clust,
345 trash_init, trash_fini,
349 UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
350 uma_zone_set_allocf(zone_jumbo9, mbuf_jumbo_alloc);
352 nmbjumbo9 = uma_zone_set_max(zone_jumbo9, nmbjumbo9);
353 uma_zone_set_warning(zone_jumbo9, "kern.ipc.nmbjumbo9 limit reached");
355 zone_jumbo16 = uma_zcreate(MBUF_JUMBO16_MEM_NAME, MJUM16BYTES,
356 mb_ctor_clust, mb_dtor_clust,
358 trash_init, trash_fini,
362 UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
363 uma_zone_set_allocf(zone_jumbo16, mbuf_jumbo_alloc);
365 nmbjumbo16 = uma_zone_set_max(zone_jumbo16, nmbjumbo16);
366 uma_zone_set_warning(zone_jumbo16, "kern.ipc.nmbjumbo16 limit reached");
368 zone_ext_refcnt = uma_zcreate(MBUF_EXTREFCNT_MEM_NAME, sizeof(u_int),
371 UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
373 /* uma_prealloc() goes here... */
376 * Hook event handler for low-memory situation, used to
377 * drain protocols and push data back to the caches (UMA
378 * later pushes it back to VM).
380 EVENTHANDLER_REGISTER(vm_lowmem, mb_reclaim, NULL,
381 EVENTHANDLER_PRI_FIRST);
383 SYSINIT(mbuf, SI_SUB_MBUF, SI_ORDER_FIRST, mbuf_init, NULL);
386 * UMA backend page allocator for the jumbo frame zones.
388 * Allocates kernel virtual memory that is backed by contiguous physical
392 mbuf_jumbo_alloc(uma_zone_t zone, int bytes, uint8_t *flags, int wait)
395 /* Inform UMA that this allocator uses kernel_map/object. */
396 *flags = UMA_SLAB_KERNEL;
397 return ((void *)kmem_alloc_contig(kernel_arena, bytes, wait,
398 (vm_paddr_t)0, ~(vm_paddr_t)0, 1, 0, VM_MEMATTR_DEFAULT));
402 * Constructor for Mbuf master zone.
404 * The 'arg' pointer points to a mb_args structure which
405 * contains call-specific information required to support the
406 * mbuf allocation API. See mbuf.h.
409 mb_ctor_mbuf(void *mem, int size, void *arg, int how)
412 struct mb_args *args;
418 trash_ctor(mem, size, arg, how);
420 args = (struct mb_args *)arg;
424 * The mbuf is initialized later. The caller has the
425 * responsibility to set up any MAC labels too.
427 if (type == MT_NOINIT)
430 m = (struct mbuf *)mem;
433 error = m_init(m, NULL, size, how, type, flags);
439 * The Mbuf master zone destructor.
442 mb_dtor_mbuf(void *mem, int size, void *arg)
447 m = (struct mbuf *)mem;
448 flags = (unsigned long)arg;
450 if ((m->m_flags & M_PKTHDR) && !SLIST_EMPTY(&m->m_pkthdr.tags))
451 m_tag_delete_chain(m, NULL);
452 KASSERT((m->m_flags & M_EXT) == 0, ("%s: M_EXT set", __func__));
453 KASSERT((m->m_flags & M_NOFREE) == 0, ("%s: M_NOFREE set", __func__));
455 trash_dtor(mem, size, arg);
460 * The Mbuf Packet zone destructor.
463 mb_dtor_pack(void *mem, int size, void *arg)
467 m = (struct mbuf *)mem;
468 if ((m->m_flags & M_PKTHDR) != 0)
469 m_tag_delete_chain(m, NULL);
471 /* Make sure we've got a clean cluster back. */
472 KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__));
473 KASSERT(m->m_ext.ext_buf != NULL, ("%s: ext_buf == NULL", __func__));
474 KASSERT(m->m_ext.ext_free == NULL, ("%s: ext_free != NULL", __func__));
475 KASSERT(m->m_ext.ext_arg1 == NULL, ("%s: ext_arg1 != NULL", __func__));
476 KASSERT(m->m_ext.ext_arg2 == NULL, ("%s: ext_arg2 != NULL", __func__));
477 KASSERT(m->m_ext.ext_size == MCLBYTES, ("%s: ext_size != MCLBYTES", __func__));
478 KASSERT(m->m_ext.ext_type == EXT_PACKET, ("%s: ext_type != EXT_PACKET", __func__));
479 KASSERT(*m->m_ext.ref_cnt == 1, ("%s: ref_cnt != 1", __func__));
481 trash_dtor(m->m_ext.ext_buf, MCLBYTES, arg);
484 * If there are processes blocked on zone_clust, waiting for pages
485 * to be freed up, * cause them to be woken up by draining the
486 * packet zone. We are exposed to a race here * (in the check for
487 * the UMA_ZFLAG_FULL) where we might miss the flag set, but that
488 * is deliberate. We don't want to acquire the zone lock for every
491 if (uma_zone_exhausted_nolock(zone_clust))
492 zone_drain(zone_pack);
496 * The Cluster and Jumbo[PAGESIZE|9|16] zone constructor.
498 * Here the 'arg' pointer points to the Mbuf which we
499 * are configuring cluster storage for. If 'arg' is
500 * empty we allocate just the cluster without setting
501 * the mbuf to it. See mbuf.h.
504 mb_ctor_clust(void *mem, int size, void *arg, int how)
512 trash_ctor(mem, size, arg, how);
519 #if MJUMPAGESIZE != MCLBYTES
534 panic("unknown cluster size");
538 m = (struct mbuf *)arg;
539 refcnt = uma_find_refcnt(zone, mem);
542 m->m_ext.ext_buf = (caddr_t)mem;
543 m->m_data = m->m_ext.ext_buf;
545 m->m_ext.ext_free = NULL;
546 m->m_ext.ext_arg1 = NULL;
547 m->m_ext.ext_arg2 = NULL;
548 m->m_ext.ext_size = size;
549 m->m_ext.ext_type = type;
550 m->m_ext.ext_flags = 0;
551 m->m_ext.ref_cnt = refcnt;
558 * The Mbuf Cluster zone destructor.
561 mb_dtor_clust(void *mem, int size, void *arg)
566 zone = m_getzone(size);
567 KASSERT(*(uma_find_refcnt(zone, mem)) <= 1,
568 ("%s: refcnt incorrect %u", __func__,
569 *(uma_find_refcnt(zone, mem))) );
571 trash_dtor(mem, size, arg);
576 * The Packet secondary zone's init routine, executed on the
577 * object's transition from mbuf keg slab to zone cache.
580 mb_zinit_pack(void *mem, int size, int how)
584 m = (struct mbuf *)mem; /* m is virgin. */
585 if (uma_zalloc_arg(zone_clust, m, how) == NULL ||
586 m->m_ext.ext_buf == NULL)
588 m->m_ext.ext_type = EXT_PACKET; /* Override. */
590 trash_init(m->m_ext.ext_buf, MCLBYTES, how);
596 * The Packet secondary zone's fini routine, executed on the
597 * object's transition from zone cache to keg slab.
600 mb_zfini_pack(void *mem, int size)
604 m = (struct mbuf *)mem;
606 trash_fini(m->m_ext.ext_buf, MCLBYTES);
608 uma_zfree_arg(zone_clust, m->m_ext.ext_buf, NULL);
610 trash_dtor(mem, size, NULL);
615 * The "packet" keg constructor.
618 mb_ctor_pack(void *mem, int size, void *arg, int how)
621 struct mb_args *args;
625 m = (struct mbuf *)mem;
626 args = (struct mb_args *)arg;
631 trash_ctor(m->m_ext.ext_buf, MCLBYTES, arg, how);
634 error = m_init(m, NULL, size, how, type, flags);
636 /* m_ext is already initialized. */
637 m->m_data = m->m_ext.ext_buf;
638 m->m_flags = (flags | M_EXT);
644 m_pkthdr_init(struct mbuf *m, int how)
649 m->m_data = m->m_pktdat;
650 m->m_pkthdr.rcvif = NULL;
651 SLIST_INIT(&m->m_pkthdr.tags);
653 m->m_pkthdr.flowid = 0;
654 m->m_pkthdr.csum_flags = 0;
655 m->m_pkthdr.fibnum = 0;
656 m->m_pkthdr.cosqos = 0;
657 m->m_pkthdr.rsstype = 0;
658 m->m_pkthdr.l2hlen = 0;
659 m->m_pkthdr.l3hlen = 0;
660 m->m_pkthdr.l4hlen = 0;
661 m->m_pkthdr.l5hlen = 0;
662 m->m_pkthdr.PH_per.sixtyfour[0] = 0;
663 m->m_pkthdr.PH_loc.sixtyfour[0] = 0;
665 /* If the label init fails, fail the alloc */
666 error = mac_mbuf_init(m, how);
675 * This is the protocol drain routine.
677 * No locks should be held when this is called. The drain routines have to
678 * presently acquire some locks which raises the possibility of lock order
682 mb_reclaim(void *junk)
687 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK | WARN_PANIC, NULL,
690 for (dp = domains; dp != NULL; dp = dp->dom_next)
691 for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++)
692 if (pr->pr_drain != NULL)