xref: /freebsd/sys/kern/kern_mbuf.c (revision 3642298923e528d795e3a30ec165d2b469e28b40)
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_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 #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 int nmbclusters;
84 struct mbstat mbstat;
85 
86 static void
87 tunable_mbinit(void *dummy)
88 {
89 
90 	/* This has to be done before VM init. */
91 	nmbclusters = 1024 + maxusers * 64;
92 	TUNABLE_INT_FETCH("kern.ipc.nmbclusters", &nmbclusters);
93 }
94 SYSINIT(tunable_mbinit, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_mbinit, NULL);
95 
96 SYSCTL_DECL(_kern_ipc);
97 SYSCTL_INT(_kern_ipc, OID_AUTO, nmbclusters, CTLFLAG_RW, &nmbclusters, 0,
98     "Maximum number of mbuf clusters allowed");
99 SYSCTL_STRUCT(_kern_ipc, OID_AUTO, mbstat, CTLFLAG_RD, &mbstat, mbstat,
100     "Mbuf general information and statistics");
101 
102 /*
103  * Zones from which we allocate.
104  */
105 uma_zone_t	zone_mbuf;
106 uma_zone_t	zone_clust;
107 uma_zone_t	zone_pack;
108 
109 /*
110  * Local prototypes.
111  */
112 static int	mb_ctor_mbuf(void *, int, void *, int);
113 static int	mb_ctor_clust(void *, int, void *, int);
114 static int	mb_ctor_pack(void *, int, void *, int);
115 static void	mb_dtor_mbuf(void *, int, void *);
116 static void	mb_dtor_clust(void *, int, void *);	/* XXX */
117 static void	mb_dtor_pack(void *, int, void *);	/* XXX */
118 static int	mb_init_pack(void *, int, int);
119 static void	mb_fini_pack(void *, int);
120 
121 static void	mb_reclaim(void *);
122 static void	mbuf_init(void *);
123 
124 /* Ensure that MSIZE doesn't break dtom() - it must be a power of 2 */
125 CTASSERT((((MSIZE - 1) ^ MSIZE) + 1) >> 1 == MSIZE);
126 
127 /*
128  * Initialize FreeBSD Network buffer allocation.
129  */
130 SYSINIT(mbuf, SI_SUB_MBUF, SI_ORDER_FIRST, mbuf_init, NULL)
131 static void
132 mbuf_init(void *dummy)
133 {
134 
135 	/*
136 	 * Configure UMA zones for Mbufs, Clusters, and Packets.
137 	 */
138 	zone_mbuf = uma_zcreate(MBUF_MEM_NAME, MSIZE, mb_ctor_mbuf,
139 	    mb_dtor_mbuf,
140 #ifdef INVARIANTS
141 	    trash_init, trash_fini, MSIZE - 1, UMA_ZONE_MAXBUCKET);
142 #else
143 	    NULL, NULL, MSIZE - 1, UMA_ZONE_MAXBUCKET);
144 #endif
145 	zone_clust = uma_zcreate(MBUF_CLUSTER_MEM_NAME, MCLBYTES,
146 	    mb_ctor_clust,
147 #ifdef INVARIANTS
148 	    mb_dtor_clust, trash_init, trash_fini, UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
149 #else
150 	    mb_dtor_clust, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_REFCNT);
151 #endif
152 	if (nmbclusters > 0)
153 		uma_zone_set_max(zone_clust, nmbclusters);
154 	zone_pack = uma_zsecond_create(MBUF_PACKET_MEM_NAME, mb_ctor_pack,
155 	    mb_dtor_pack, mb_init_pack, mb_fini_pack, zone_mbuf);
156 
157 	/* uma_prealloc() goes here */
158 
159 	/*
160 	 * Hook event handler for low-memory situation, used to
161 	 * drain protocols and push data back to the caches (UMA
162 	 * later pushes it back to VM).
163 	 */
164 	EVENTHANDLER_REGISTER(vm_lowmem, mb_reclaim, NULL,
165 	    EVENTHANDLER_PRI_FIRST);
166 
167 	/*
168 	 * [Re]set counters and local statistics knobs.
169 	 * XXX Some of these should go and be replaced, but UMA stat
170 	 * gathering needs to be revised.
171 	 */
172 	mbstat.m_mbufs = 0;
173 	mbstat.m_mclusts = 0;
174 	mbstat.m_drain = 0;
175 	mbstat.m_msize = MSIZE;
176 	mbstat.m_mclbytes = MCLBYTES;
177 	mbstat.m_minclsize = MINCLSIZE;
178 	mbstat.m_mlen = MLEN;
179 	mbstat.m_mhlen = MHLEN;
180 	mbstat.m_numtypes = MT_NTYPES;
181 
182 	mbstat.m_mcfail = mbstat.m_mpfail = 0;
183 	mbstat.sf_iocnt = 0;
184 	mbstat.sf_allocwait = mbstat.sf_allocfail = 0;
185 }
186 
187 /*
188  * Constructor for Mbuf master zone.
189  *
190  * The 'arg' pointer points to a mb_args structure which
191  * contains call-specific information required to support the
192  * mbuf allocation API.
193  */
194 static int
195 mb_ctor_mbuf(void *mem, int size, void *arg, int how)
196 {
197 	struct mbuf *m;
198 	struct mb_args *args;
199 #ifdef MAC
200 	int error;
201 #endif
202 	int flags;
203 	short type;
204 
205 #ifdef INVARIANTS
206 	trash_ctor(mem, size, arg, how);
207 #endif
208 	m = (struct mbuf *)mem;
209 	args = (struct mb_args *)arg;
210 	flags = args->flags;
211 	type = args->type;
212 
213 	m->m_type = type;
214 	m->m_next = NULL;
215 	m->m_nextpkt = NULL;
216 	m->m_flags = flags;
217 	if (flags & M_PKTHDR) {
218 		m->m_data = m->m_pktdat;
219 		m->m_pkthdr.rcvif = NULL;
220 		m->m_pkthdr.csum_flags = 0;
221 		SLIST_INIT(&m->m_pkthdr.tags);
222 #ifdef MAC
223 		/* If the label init fails, fail the alloc */
224 		error = mac_init_mbuf(m, how);
225 		if (error)
226 			return (error);
227 #endif
228 	} else
229 		m->m_data = m->m_dat;
230 	mbstat.m_mbufs += 1;	/* XXX */
231 	return (0);
232 }
233 
234 /*
235  * The Mbuf master zone and Packet secondary zone destructor.
236  */
237 static void
238 mb_dtor_mbuf(void *mem, int size, void *arg)
239 {
240 	struct mbuf *m;
241 
242 	m = (struct mbuf *)mem;
243 	if ((m->m_flags & M_PKTHDR) != 0)
244 		m_tag_delete_chain(m, NULL);
245 #ifdef INVARIANTS
246 	trash_dtor(mem, size, arg);
247 #endif
248 	mbstat.m_mbufs -= 1;	/* XXX */
249 }
250 
251 /* XXX Only because of stats */
252 static void
253 mb_dtor_pack(void *mem, int size, void *arg)
254 {
255 	struct mbuf *m;
256 
257 	m = (struct mbuf *)mem;
258 	if ((m->m_flags & M_PKTHDR) != 0)
259 		m_tag_delete_chain(m, NULL);
260 #ifdef INVARIANTS
261 	trash_dtor(m->m_ext.ext_buf, MCLBYTES, arg);
262 #endif
263 	mbstat.m_mbufs -= 1;	/* XXX */
264 	mbstat.m_mclusts -= 1;	/* XXX */
265 }
266 
267 /*
268  * The Cluster zone constructor.
269  *
270  * Here the 'arg' pointer points to the Mbuf which we
271  * are configuring cluster storage for.
272  */
273 static int
274 mb_ctor_clust(void *mem, int size, void *arg, int how)
275 {
276 	struct mbuf *m;
277 
278 #ifdef INVARIANTS
279 	trash_ctor(mem, size, arg, how);
280 #endif
281 	m = (struct mbuf *)arg;
282 	m->m_ext.ext_buf = (caddr_t)mem;
283 	m->m_data = m->m_ext.ext_buf;
284 	m->m_flags |= M_EXT;
285 	m->m_ext.ext_free = NULL;
286 	m->m_ext.ext_args = NULL;
287 	m->m_ext.ext_size = MCLBYTES;
288 	m->m_ext.ext_type = EXT_CLUSTER;
289 	m->m_ext.ref_cnt = NULL;	/* Lazy counter assign. */
290 	mbstat.m_mclusts += 1;	/* XXX */
291 	return (0);
292 }
293 
294 /* XXX */
295 static void
296 mb_dtor_clust(void *mem, int size, void *arg)
297 {
298 #ifdef INVARIANTS
299 	trash_dtor(mem, size, arg);
300 #endif
301 	mbstat.m_mclusts -= 1;	/* XXX */
302 }
303 
304 /*
305  * The Packet secondary zone's init routine, executed on the
306  * object's transition from keg slab to zone cache.
307  */
308 static int
309 mb_init_pack(void *mem, int size, int how)
310 {
311 	struct mbuf *m;
312 
313 	m = (struct mbuf *)mem;
314 	m->m_ext.ext_buf = NULL;
315 	uma_zalloc_arg(zone_clust, m, how);
316 	if (m->m_ext.ext_buf == NULL)
317 		return (ENOMEM);
318 #ifdef INVARIANTS
319 	trash_init(m->m_ext.ext_buf, MCLBYTES, how);
320 #endif
321 	mbstat.m_mclusts -= 1;	/* XXX */
322 	return (0);
323 }
324 
325 /*
326  * The Packet secondary zone's fini routine, executed on the
327  * object's transition from zone cache to keg slab.
328  */
329 static void
330 mb_fini_pack(void *mem, int size)
331 {
332 	struct mbuf *m;
333 
334 	m = (struct mbuf *)mem;
335 #ifdef INVARIANTS
336 	trash_fini(m->m_ext.ext_buf, MCLBYTES);
337 #endif
338 	uma_zfree_arg(zone_clust, m->m_ext.ext_buf, NULL);
339 	m->m_ext.ext_buf = NULL;
340 	mbstat.m_mclusts += 1;	/* XXX */
341 #ifdef INVARIANTS
342 	trash_dtor(mem, size, NULL);
343 #endif
344 }
345 
346 /*
347  * The "packet" keg constructor.
348  */
349 static int
350 mb_ctor_pack(void *mem, int size, void *arg, int how)
351 {
352 	struct mbuf *m;
353 	struct mb_args *args;
354 #ifdef MAC
355 	int error;
356 #endif
357 	int flags;
358 	short type;
359 
360 	m = (struct mbuf *)mem;
361 	args = (struct mb_args *)arg;
362 	flags = args->flags;
363 	type = args->type;
364 
365 #ifdef INVARIANTS
366 	trash_ctor(m->m_ext.ext_buf, MCLBYTES, arg, how);
367 #endif
368 	m->m_type = type;
369 	m->m_next = NULL;
370 	m->m_nextpkt = NULL;
371 	m->m_data = m->m_ext.ext_buf;
372 	m->m_flags = flags|M_EXT;
373 	m->m_ext.ext_free = NULL;
374 	m->m_ext.ext_args = NULL;
375 	m->m_ext.ext_size = MCLBYTES;
376 	m->m_ext.ext_type = EXT_PACKET;
377 	m->m_ext.ref_cnt = NULL;	/* Lazy counter assign. */
378 
379 	if (flags & M_PKTHDR) {
380 		m->m_pkthdr.rcvif = NULL;
381 		m->m_pkthdr.csum_flags = 0;
382 		SLIST_INIT(&m->m_pkthdr.tags);
383 #ifdef MAC
384 		/* If the label init fails, fail the alloc */
385 		error = mac_init_mbuf(m, how);
386 		if (error)
387 			return (error);
388 #endif
389 	}
390 	mbstat.m_mbufs += 1;	/* XXX */
391 	mbstat.m_mclusts += 1;	/* XXX */
392 	return (0);
393 }
394 
395 /*
396  * This is the protocol drain routine.
397  *
398  * No locks should be held when this is called.  The drain routines have to
399  * presently acquire some locks which raises the possibility of lock order
400  * reversal.
401  */
402 static void
403 mb_reclaim(void *junk)
404 {
405 	struct domain *dp;
406 	struct protosw *pr;
407 
408 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK | WARN_PANIC, NULL,
409 	    "mb_reclaim()");
410 
411 	mbstat.m_drain++;
412 	for (dp = domains; dp != NULL; dp = dp->dom_next)
413 		for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++)
414 			if (pr->pr_drain != NULL)
415 				(*pr->pr_drain)();
416 }
417