xref: /freebsd/sys/kern/kern_mbuf.c (revision 6af83ee0d2941d18880b6aaa2b4facd1d30c6106)
1 /*-
2  * Copyright (c) 2004, 2005
3  * 	Bosko Milekic <bmilekic@FreeBSD.org>.
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_mac.h"
32 #include "opt_param.h"
33 
34 #include <sys/param.h>
35 #include <sys/mac.h>
36 #include <sys/malloc.h>
37 #include <sys/systm.h>
38 #include <sys/mbuf.h>
39 #include <sys/domain.h>
40 #include <sys/eventhandler.h>
41 #include <sys/kernel.h>
42 #include <sys/protosw.h>
43 #include <sys/smp.h>
44 #include <sys/sysctl.h>
45 
46 #include <vm/vm.h>
47 #include <vm/vm_page.h>
48 #include <vm/uma.h>
49 
50 /*
51  * In FreeBSD, Mbufs and Mbuf Clusters are allocated from UMA
52  * Zones.
53  *
54  * Mbuf Clusters (2K, contiguous) are allocated from the Cluster
55  * Zone.  The Zone can be capped at kern.ipc.nmbclusters, if the
56  * administrator so desires.
57  *
58  * Mbufs are allocated from a UMA Master Zone called the Mbuf
59  * Zone.
60  *
61  * Additionally, FreeBSD provides a Packet Zone, which it
62  * configures as a Secondary Zone to the Mbuf Master Zone,
63  * thus sharing backend Slab kegs with the Mbuf Master Zone.
64  *
65  * Thus common-case allocations and locking are simplified:
66  *
67  *  m_clget()                m_getcl()
68  *    |                         |
69  *    |   .------------>[(Packet Cache)]    m_get(), m_gethdr()
70  *    |   |             [     Packet   ]            |
71  *  [(Cluster Cache)]   [    Secondary ]   [ (Mbuf Cache)     ]
72  *  [ Cluster Zone  ]   [     Zone     ]   [ Mbuf Master Zone ]
73  *        |                       \________         |
74  *  [ Cluster Keg   ]                      \       /
75  *        |    	                         [ Mbuf Keg   ]
76  *  [ Cluster Slabs ]                         |
77  *        |                              [ Mbuf Slabs ]
78  *         \____________(VM)_________________/
79  */
80 
81 int nmbclusters;
82 struct mbstat mbstat;
83 
84 static void
85 tunable_mbinit(void *dummy)
86 {
87 
88 	/* This has to be done before VM init. */
89 	nmbclusters = 1024 + maxusers * 64;
90 	TUNABLE_INT_FETCH("kern.ipc.nmbclusters", &nmbclusters);
91 }
92 SYSINIT(tunable_mbinit, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_mbinit, NULL);
93 
94 SYSCTL_DECL(_kern_ipc);
95 SYSCTL_INT(_kern_ipc, OID_AUTO, nmbclusters, CTLFLAG_RW, &nmbclusters, 0,
96     "Maximum number of mbuf clusters allowed");
97 SYSCTL_STRUCT(_kern_ipc, OID_AUTO, mbstat, CTLFLAG_RD, &mbstat, mbstat,
98     "Mbuf general information and statistics");
99 
100 /*
101  * Zones from which we allocate.
102  */
103 uma_zone_t	zone_mbuf;
104 uma_zone_t	zone_clust;
105 uma_zone_t	zone_pack;
106 
107 /*
108  * Local prototypes.
109  */
110 static int	mb_ctor_mbuf(void *, int, void *, int);
111 static int	mb_ctor_clust(void *, int, void *, int);
112 static int	mb_ctor_pack(void *, int, void *, int);
113 static void	mb_dtor_mbuf(void *, int, void *);
114 static void	mb_dtor_clust(void *, int, void *);	/* XXX */
115 static void	mb_dtor_pack(void *, int, void *);	/* XXX */
116 static int	mb_init_pack(void *, int, int);
117 static void	mb_fini_pack(void *, int);
118 
119 static void	mb_reclaim(void *);
120 static void	mbuf_init(void *);
121 
122 /* Ensure that MSIZE doesn't break dtom() - it must be a power of 2 */
123 CTASSERT((((MSIZE - 1) ^ MSIZE) + 1) >> 1 == MSIZE);
124 
125 /*
126  * Initialize FreeBSD Network buffer allocation.
127  */
128 SYSINIT(mbuf, SI_SUB_MBUF, SI_ORDER_FIRST, mbuf_init, NULL)
129 static void
130 mbuf_init(void *dummy)
131 {
132 
133 	/*
134 	 * Configure UMA zones for Mbufs, Clusters, and Packets.
135 	 */
136 	zone_mbuf = uma_zcreate("Mbuf", MSIZE, mb_ctor_mbuf, mb_dtor_mbuf,
137 	    NULL, NULL, MSIZE - 1, UMA_ZONE_MAXBUCKET);
138 	zone_clust = uma_zcreate("MbufClust", MCLBYTES, mb_ctor_clust,
139 	    mb_dtor_clust, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
140 	if (nmbclusters > 0)
141 		uma_zone_set_max(zone_clust, nmbclusters);
142 	zone_pack = uma_zsecond_create("Packet", mb_ctor_pack, mb_dtor_pack,
143 	    mb_init_pack, mb_fini_pack, zone_mbuf);
144 
145 	/* uma_prealloc() goes here */
146 
147 	/*
148 	 * Hook event handler for low-memory situation, used to
149 	 * drain protocols and push data back to the caches (UMA
150 	 * later pushes it back to VM).
151 	 */
152 	EVENTHANDLER_REGISTER(vm_lowmem, mb_reclaim, NULL,
153 	    EVENTHANDLER_PRI_FIRST);
154 
155 	/*
156 	 * [Re]set counters and local statistics knobs.
157 	 * XXX Some of these should go and be replaced, but UMA stat
158 	 * gathering needs to be revised.
159 	 */
160 	mbstat.m_mbufs = 0;
161 	mbstat.m_mclusts = 0;
162 	mbstat.m_drain = 0;
163 	mbstat.m_msize = MSIZE;
164 	mbstat.m_mclbytes = MCLBYTES;
165 	mbstat.m_minclsize = MINCLSIZE;
166 	mbstat.m_mlen = MLEN;
167 	mbstat.m_mhlen = MHLEN;
168 	mbstat.m_numtypes = MT_NTYPES;
169 
170 	mbstat.m_mcfail = mbstat.m_mpfail = 0;
171 	mbstat.sf_iocnt = 0;
172 	mbstat.sf_allocwait = mbstat.sf_allocfail = 0;
173 }
174 
175 /*
176  * Constructor for Mbuf master zone.
177  *
178  * The 'arg' pointer points to a mb_args structure which
179  * contains call-specific information required to support the
180  * mbuf allocation API.
181  */
182 static int
183 mb_ctor_mbuf(void *mem, int size, void *arg, int how)
184 {
185 	struct mbuf *m;
186 	struct mb_args *args;
187 #ifdef MAC
188 	int error;
189 #endif
190 	int flags;
191 	short type;
192 
193 	m = (struct mbuf *)mem;
194 	args = (struct mb_args *)arg;
195 	flags = args->flags;
196 	type = args->type;
197 
198 	m->m_type = type;
199 	m->m_next = NULL;
200 	m->m_nextpkt = NULL;
201 	m->m_flags = flags;
202 	if (flags & M_PKTHDR) {
203 		m->m_data = m->m_pktdat;
204 		m->m_pkthdr.rcvif = NULL;
205 		m->m_pkthdr.csum_flags = 0;
206 		SLIST_INIT(&m->m_pkthdr.tags);
207 #ifdef MAC
208 		/* If the label init fails, fail the alloc */
209 		error = mac_init_mbuf(m, how);
210 		if (error)
211 			return (error);
212 #endif
213 	} else
214 		m->m_data = m->m_dat;
215 	mbstat.m_mbufs += 1;	/* XXX */
216 	return (0);
217 }
218 
219 /*
220  * The Mbuf master zone and Packet secondary zone destructor.
221  */
222 static void
223 mb_dtor_mbuf(void *mem, int size, void *arg)
224 {
225 	struct mbuf *m;
226 
227 	m = (struct mbuf *)mem;
228 	if ((m->m_flags & M_PKTHDR) != 0)
229 		m_tag_delete_chain(m, NULL);
230 	mbstat.m_mbufs -= 1;	/* XXX */
231 }
232 
233 /* XXX Only because of stats */
234 static void
235 mb_dtor_pack(void *mem, int size, void *arg)
236 {
237 	struct mbuf *m;
238 
239 	m = (struct mbuf *)mem;
240 	if ((m->m_flags & M_PKTHDR) != 0)
241 		m_tag_delete_chain(m, NULL);
242 	mbstat.m_mbufs -= 1;	/* XXX */
243 	mbstat.m_mclusts -= 1;	/* XXX */
244 }
245 
246 /*
247  * The Cluster zone constructor.
248  *
249  * Here the 'arg' pointer points to the Mbuf which we
250  * are configuring cluster storage for.
251  */
252 static int
253 mb_ctor_clust(void *mem, int size, void *arg, int how)
254 {
255 	struct mbuf *m;
256 
257 	m = (struct mbuf *)arg;
258 	m->m_ext.ext_buf = (caddr_t)mem;
259 	m->m_data = m->m_ext.ext_buf;
260 	m->m_flags |= M_EXT;
261 	m->m_ext.ext_free = NULL;
262 	m->m_ext.ext_args = NULL;
263 	m->m_ext.ext_size = MCLBYTES;
264 	m->m_ext.ext_type = EXT_CLUSTER;
265 	m->m_ext.ref_cnt = NULL;	/* Lazy counter assign. */
266 	mbstat.m_mclusts += 1;	/* XXX */
267 	return (0);
268 }
269 
270 /* XXX */
271 static void
272 mb_dtor_clust(void *mem, int size, void *arg)
273 {
274 	mbstat.m_mclusts -= 1;	/* XXX */
275 }
276 
277 /*
278  * The Packet secondary zone's init routine, executed on the
279  * object's transition from keg slab to zone cache.
280  */
281 static int
282 mb_init_pack(void *mem, int size, int how)
283 {
284 	struct mbuf *m;
285 
286 	m = (struct mbuf *)mem;
287 	m->m_ext.ext_buf = NULL;
288 	uma_zalloc_arg(zone_clust, m, how);
289 	if (m->m_ext.ext_buf == NULL)
290 		return (ENOMEM);
291 	mbstat.m_mclusts -= 1;	/* XXX */
292 	return (0);
293 }
294 
295 /*
296  * The Packet secondary zone's fini routine, executed on the
297  * object's transition from zone cache to keg slab.
298  */
299 static void
300 mb_fini_pack(void *mem, int size)
301 {
302 	struct mbuf *m;
303 
304 	m = (struct mbuf *)mem;
305 	uma_zfree_arg(zone_clust, m->m_ext.ext_buf, NULL);
306 	m->m_ext.ext_buf = NULL;
307 	mbstat.m_mclusts += 1;	/* XXX */
308 }
309 
310 /*
311  * The "packet" keg constructor.
312  */
313 static int
314 mb_ctor_pack(void *mem, int size, void *arg, int how)
315 {
316 	struct mbuf *m;
317 	struct mb_args *args;
318 #ifdef MAC
319 	int error;
320 #endif
321 	int flags;
322 	short type;
323 
324 	m = (struct mbuf *)mem;
325 	args = (struct mb_args *)arg;
326 	flags = args->flags;
327 	type = args->type;
328 
329 	m->m_type = type;
330 	m->m_next = NULL;
331 	m->m_nextpkt = NULL;
332 	m->m_data = m->m_ext.ext_buf;
333 	m->m_flags = flags|M_EXT;
334 	m->m_ext.ext_free = NULL;
335 	m->m_ext.ext_args = NULL;
336 	m->m_ext.ext_size = MCLBYTES;
337 	m->m_ext.ext_type = EXT_PACKET;
338 	m->m_ext.ref_cnt = NULL;	/* Lazy counter assign. */
339 
340 	if (flags & M_PKTHDR) {
341 		m->m_pkthdr.rcvif = NULL;
342 		m->m_pkthdr.csum_flags = 0;
343 		SLIST_INIT(&m->m_pkthdr.tags);
344 #ifdef MAC
345 		/* If the label init fails, fail the alloc */
346 		error = mac_init_mbuf(m, how);
347 		if (error)
348 			return (error);
349 #endif
350 	}
351 	mbstat.m_mbufs += 1;	/* XXX */
352 	mbstat.m_mclusts += 1;	/* XXX */
353 	return (0);
354 }
355 
356 /*
357  * This is the protocol drain routine.
358  *
359  * No locks should be held when this is called.  The drain routines have to
360  * presently acquire some locks which raises the possibility of lock order
361  * reversal.
362  */
363 static void
364 mb_reclaim(void *junk)
365 {
366 	struct domain *dp;
367 	struct protosw *pr;
368 
369 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK | WARN_PANIC, NULL,
370 	    "mb_reclaim()");
371 
372 	mbstat.m_drain++;
373 	for (dp = domains; dp != NULL; dp = dp->dom_next)
374 		for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++)
375 			if (pr->pr_drain != NULL)
376 				(*pr->pr_drain)();
377 }
378