xref: /freebsd/sys/kern/kern_mbuf.c (revision 63f9a4cb2684a303e3eb2ffed39c03a2e2b28ae0)
1 /*-
2  * Copyright (c) 2004
3  * 	Bosko Milekic <bmilekic@FreeBSD.org>.
4  *	All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice unmodified, this list of conditions and the following
11  *    disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the author nor the names of contributors may be
16  *    used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_mac.h"
36 #include "opt_param.h"
37 
38 #include <sys/param.h>
39 #include <sys/mac.h>
40 #include <sys/malloc.h>
41 #include <sys/systm.h>
42 #include <sys/mbuf.h>
43 #include <sys/domain.h>
44 #include <sys/eventhandler.h>
45 #include <sys/kernel.h>
46 #include <sys/protosw.h>
47 #include <sys/smp.h>
48 #include <sys/sysctl.h>
49 
50 #include <vm/vm.h>
51 #include <vm/vm_page.h>
52 #include <vm/uma.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 int nmbclusters;
86 struct mbstat mbstat;
87 
88 static void
89 tunable_mbinit(void *dummy)
90 {
91 
92 	/* This has to be done before VM init. */
93 	nmbclusters = 1024 + maxusers * 64;
94 	TUNABLE_INT_FETCH("kern.ipc.nmbclusters", &nmbclusters);
95 }
96 SYSINIT(tunable_mbinit, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_mbinit, NULL);
97 
98 SYSCTL_DECL(_kern_ipc);
99 SYSCTL_INT(_kern_ipc, OID_AUTO, nmbclusters, CTLFLAG_RW, &nmbclusters, 0,
100     "Maximum number of mbuf clusters allowed");
101 SYSCTL_STRUCT(_kern_ipc, OID_AUTO, mbstat, CTLFLAG_RD, &mbstat, mbstat,
102     "Mbuf general information and statistics");
103 
104 /*
105  * Zones from which we allocate.
106  */
107 uma_zone_t	zone_mbuf;
108 uma_zone_t	zone_clust;
109 uma_zone_t	zone_pack;
110 
111 /*
112  * Local prototypes.
113  */
114 static int	mb_ctor_mbuf(void *, int, void *, int);
115 static int	mb_ctor_clust(void *, int, void *, int);
116 static int	mb_ctor_pack(void *, int, void *, int);
117 static void	mb_dtor_mbuf(void *, int, void *);
118 static void	mb_dtor_clust(void *, int, void *);	/* XXX */
119 static void	mb_dtor_pack(void *, int, void *);	/* XXX */
120 static int	mb_init_pack(void *, int, int);
121 static void	mb_fini_pack(void *, int);
122 
123 static void	mb_reclaim(void *);
124 static void	mbuf_init(void *);
125 
126 /* Ensure that MSIZE doesn't break dtom() - it must be a power of 2 */
127 CTASSERT((((MSIZE - 1) ^ MSIZE) + 1) >> 1 == MSIZE);
128 
129 /*
130  * Initialize FreeBSD Network buffer allocation.
131  */
132 SYSINIT(mbuf, SI_SUB_MBUF, SI_ORDER_FIRST, mbuf_init, NULL)
133 static void
134 mbuf_init(void *dummy)
135 {
136 
137 	/*
138 	 * Configure UMA zones for Mbufs, Clusters, and Packets.
139 	 */
140 	zone_mbuf = uma_zcreate("Mbuf", MSIZE, mb_ctor_mbuf, mb_dtor_mbuf,
141 	    NULL, NULL, MSIZE - 1, UMA_ZONE_MAXBUCKET);
142 	zone_clust = uma_zcreate("MbufClust", MCLBYTES, mb_ctor_clust,
143 	    mb_dtor_clust, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
144 	if (nmbclusters > 0)
145 		uma_zone_set_max(zone_clust, nmbclusters);
146 	zone_pack = uma_zsecond_create("Packet", mb_ctor_pack, mb_dtor_pack,
147 	    mb_init_pack, mb_fini_pack, zone_mbuf);
148 
149 	/* uma_prealloc() goes here */
150 
151 	/*
152 	 * Hook event handler for low-memory situation, used to
153 	 * drain protocols and push data back to the caches (UMA
154 	 * later pushes it back to VM).
155 	 */
156 	EVENTHANDLER_REGISTER(vm_lowmem, mb_reclaim, NULL,
157 	    EVENTHANDLER_PRI_FIRST);
158 
159 	/*
160 	 * [Re]set counters and local statistics knobs.
161 	 * XXX Some of these should go and be replaced, but UMA stat
162 	 * gathering needs to be revised.
163 	 */
164 	mbstat.m_mbufs = 0;
165 	mbstat.m_mclusts = 0;
166 	mbstat.m_drain = 0;
167 	mbstat.m_msize = MSIZE;
168 	mbstat.m_mclbytes = MCLBYTES;
169 	mbstat.m_minclsize = MINCLSIZE;
170 	mbstat.m_mlen = MLEN;
171 	mbstat.m_mhlen = MHLEN;
172 	mbstat.m_numtypes = MT_NTYPES;
173 
174 	mbstat.m_mcfail = mbstat.m_mpfail = 0;
175 	mbstat.sf_iocnt = 0;
176 	mbstat.sf_allocwait = mbstat.sf_allocfail = 0;
177 }
178 
179 /*
180  * Constructor for Mbuf master zone.
181  *
182  * The 'arg' pointer points to a mb_args structure which
183  * contains call-specific information required to support the
184  * mbuf allocation API.
185  */
186 static int
187 mb_ctor_mbuf(void *mem, int size, void *arg, int how)
188 {
189 	struct mbuf *m;
190 	struct mb_args *args;
191 #ifdef MAC
192 	int error;
193 #endif
194 	int flags;
195 	short type;
196 
197 	m = (struct mbuf *)mem;
198 	args = (struct mb_args *)arg;
199 	flags = args->flags;
200 	type = args->type;
201 
202 	m->m_type = type;
203 	m->m_next = NULL;
204 	m->m_nextpkt = NULL;
205 	m->m_flags = flags;
206 	if (flags & M_PKTHDR) {
207 		m->m_data = m->m_pktdat;
208 		m->m_pkthdr.rcvif = NULL;
209 		m->m_pkthdr.csum_flags = 0;
210 		SLIST_INIT(&m->m_pkthdr.tags);
211 #ifdef MAC
212 		/* If the label init fails, fail the alloc */
213 		error = mac_init_mbuf(m, how);
214 		if (error)
215 			return (error);
216 #endif
217 	} else
218 		m->m_data = m->m_dat;
219 	mbstat.m_mbufs += 1;	/* XXX */
220 	return (0);
221 }
222 
223 /*
224  * The Mbuf master zone and Packet secondary zone destructor.
225  */
226 static void
227 mb_dtor_mbuf(void *mem, int size, void *arg)
228 {
229 	struct mbuf *m;
230 
231 	m = (struct mbuf *)mem;
232 	if ((m->m_flags & M_PKTHDR) != 0)
233 		m_tag_delete_chain(m, NULL);
234 	mbstat.m_mbufs -= 1;	/* XXX */
235 }
236 
237 /* XXX Only because of stats */
238 static void
239 mb_dtor_pack(void *mem, int size, void *arg)
240 {
241 	struct mbuf *m;
242 
243 	m = (struct mbuf *)mem;
244 	if ((m->m_flags & M_PKTHDR) != 0)
245 		m_tag_delete_chain(m, NULL);
246 	mbstat.m_mbufs -= 1;	/* XXX */
247 	mbstat.m_mclusts -= 1;	/* XXX */
248 }
249 
250 /*
251  * The Cluster zone constructor.
252  *
253  * Here the 'arg' pointer points to the Mbuf which we
254  * are configuring cluster storage for.
255  */
256 static int
257 mb_ctor_clust(void *mem, int size, void *arg, int how)
258 {
259 	struct mbuf *m;
260 
261 	m = (struct mbuf *)arg;
262 	m->m_ext.ext_buf = (caddr_t)mem;
263 	m->m_data = m->m_ext.ext_buf;
264 	m->m_flags |= M_EXT;
265 	m->m_ext.ext_free = NULL;
266 	m->m_ext.ext_args = NULL;
267 	m->m_ext.ext_size = MCLBYTES;
268 	m->m_ext.ext_type = EXT_CLUSTER;
269 	m->m_ext.ref_cnt = (u_int *)uma_find_refcnt(zone_clust,
270 	    m->m_ext.ext_buf);
271 	*(m->m_ext.ref_cnt) = 1;
272 	mbstat.m_mclusts += 1;	/* XXX */
273 	return (0);
274 }
275 
276 /* XXX */
277 static void
278 mb_dtor_clust(void *mem, int size, void *arg)
279 {
280 	mbstat.m_mclusts -= 1;	/* XXX */
281 }
282 
283 /*
284  * The Packet secondary zone's init routine, executed on the
285  * object's transition from keg slab to zone cache.
286  */
287 static int
288 mb_init_pack(void *mem, int size, int how)
289 {
290 	struct mbuf *m;
291 
292 	m = (struct mbuf *)mem;
293 	m->m_ext.ext_buf = NULL;
294 	uma_zalloc_arg(zone_clust, m, how);
295 	if (m->m_ext.ext_buf == NULL)
296 		return (ENOMEM);
297 	mbstat.m_mclusts -= 1;	/* XXX */
298 	return (0);
299 }
300 
301 /*
302  * The Packet secondary zone's fini routine, executed on the
303  * object's transition from zone cache to keg slab.
304  */
305 static void
306 mb_fini_pack(void *mem, int size)
307 {
308 	struct mbuf *m;
309 
310 	m = (struct mbuf *)mem;
311 	uma_zfree_arg(zone_clust, m->m_ext.ext_buf, NULL);
312 	m->m_ext.ext_buf = NULL;
313 	mbstat.m_mclusts += 1;	/* XXX */
314 }
315 
316 /*
317  * The "packet" keg constructor.
318  */
319 static int
320 mb_ctor_pack(void *mem, int size, void *arg, int how)
321 {
322 	struct mbuf *m;
323 	struct mb_args *args;
324 #ifdef MAC
325 	int error;
326 #endif
327 	int flags;
328 	short type;
329 
330 	m = (struct mbuf *)mem;
331 	args = (struct mb_args *)arg;
332 	flags = args->flags;
333 	type = args->type;
334 
335 	m->m_type = type;
336 	m->m_next = NULL;
337 	m->m_nextpkt = NULL;
338 	m->m_data = m->m_ext.ext_buf;
339 	m->m_flags = flags|M_EXT;
340 	m->m_ext.ext_free = NULL;
341 	m->m_ext.ext_args = NULL;
342 	m->m_ext.ext_size = MCLBYTES;
343 	m->m_ext.ext_type = EXT_PACKET;
344 	*(m->m_ext.ref_cnt) = 1;
345 
346 	if (flags & M_PKTHDR) {
347 		m->m_pkthdr.rcvif = NULL;
348 		m->m_pkthdr.csum_flags = 0;
349 		SLIST_INIT(&m->m_pkthdr.tags);
350 #ifdef MAC
351 		/* If the label init fails, fail the alloc */
352 		error = mac_init_mbuf(m, how);
353 		if (error)
354 			return (error);
355 #endif
356 	}
357 	mbstat.m_mbufs += 1;	/* XXX */
358 	mbstat.m_mclusts += 1;	/* XXX */
359 	return (0);
360 }
361 
362 /*
363  * This is the protocol drain routine.
364  *
365  * No locks should be held when this is called.  The drain routines have to
366  * presently acquire some locks which raises the possibility of lock order
367  * reversal.
368  */
369 static void
370 mb_reclaim(void *junk)
371 {
372 	struct domain *dp;
373 	struct protosw *pr;
374 
375 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK | WARN_PANIC, NULL,
376 	    "mb_reclaim()");
377 
378 	mbstat.m_drain++;
379 	for (dp = domains; dp != NULL; dp = dp->dom_next)
380 		for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++)
381 			if (pr->pr_drain != NULL)
382 				(*pr->pr_drain)();
383 }
384