subr_sfbuf.c revision 303975 
1/*-
2 * Copyright (c) 2014 Gleb Smirnoff <glebius@FreeBSD.org>
3 * Copyright (c) 2003, 2005 Alan L. Cox <alc@cs.rice.edu>
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, this list of conditions and the following 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: releng/11.0/sys/kern/subr_sfbuf.c 269813 2014-08-11 15:06:07Z glebius $");
30
31#include <sys/param.h>
32#include <sys/kernel.h>
33#include <sys/lock.h>
34#include <sys/malloc.h>
35#include <sys/mutex.h>
36#include <sys/sf_buf.h>
37#include <sys/smp.h>
38#include <sys/sysctl.h>
39
40#include <vm/vm.h>
41#include <vm/vm_extern.h>
42#include <vm/vm_page.h>
43
44#ifndef NSFBUFS
45#define	NSFBUFS		(512 + maxusers * 16)
46#endif
47
48static int nsfbufs;
49static int nsfbufspeak;
50static int nsfbufsused;
51
52SYSCTL_INT(_kern_ipc, OID_AUTO, nsfbufs, CTLFLAG_RDTUN, &nsfbufs, 0,
53    "Maximum number of sendfile(2) sf_bufs available");
54SYSCTL_INT(_kern_ipc, OID_AUTO, nsfbufspeak, CTLFLAG_RD, &nsfbufspeak, 0,
55    "Number of sendfile(2) sf_bufs at peak usage");
56SYSCTL_INT(_kern_ipc, OID_AUTO, nsfbufsused, CTLFLAG_RD, &nsfbufsused, 0,
57    "Number of sendfile(2) sf_bufs in use");
58
59static void	sf_buf_init(void *arg);
60SYSINIT(sock_sf, SI_SUB_MBUF, SI_ORDER_ANY, sf_buf_init, NULL);
61
62LIST_HEAD(sf_head, sf_buf);
63
64/*
65 * A hash table of active sendfile(2) buffers
66 */
67static struct sf_head *sf_buf_active;
68static u_long sf_buf_hashmask;
69
70#define	SF_BUF_HASH(m)	(((m) - vm_page_array) & sf_buf_hashmask)
71
72static TAILQ_HEAD(, sf_buf) sf_buf_freelist;
73static u_int	sf_buf_alloc_want;
74
75/*
76 * A lock used to synchronize access to the hash table and free list
77 */
78static struct mtx sf_buf_lock;
79
80/*
81 * Allocate a pool of sf_bufs (sendfile(2) or "super-fast" if you prefer. :-))
82 */
83static void
84sf_buf_init(void *arg)
85{
86	struct sf_buf *sf_bufs;
87	vm_offset_t sf_base;
88	int i;
89
90#ifdef SFBUF_OPTIONAL_DIRECT_MAP
91	if (SFBUF_OPTIONAL_DIRECT_MAP)
92		return;
93#endif
94
95	nsfbufs = NSFBUFS;
96	TUNABLE_INT_FETCH("kern.ipc.nsfbufs", &nsfbufs);
97
98	sf_buf_active = hashinit(nsfbufs, M_TEMP, &sf_buf_hashmask);
99	TAILQ_INIT(&sf_buf_freelist);
100	sf_base = kva_alloc(nsfbufs * PAGE_SIZE);
101	sf_bufs = malloc(nsfbufs * sizeof(struct sf_buf), M_TEMP,
102	    M_WAITOK | M_ZERO);
103	for (i = 0; i < nsfbufs; i++) {
104		sf_bufs[i].kva = sf_base + i * PAGE_SIZE;
105		TAILQ_INSERT_TAIL(&sf_buf_freelist, &sf_bufs[i], free_entry);
106	}
107	sf_buf_alloc_want = 0;
108	mtx_init(&sf_buf_lock, "sf_buf", NULL, MTX_DEF);
109}
110
111/*
112 * Get an sf_buf from the freelist.  May block if none are available.
113 */
114struct sf_buf *
115sf_buf_alloc(struct vm_page *m, int flags)
116{
117	struct sf_head *hash_list;
118	struct sf_buf *sf;
119	int error;
120
121#ifdef SFBUF_OPTIONAL_DIRECT_MAP
122	if (SFBUF_OPTIONAL_DIRECT_MAP)
123		return ((struct sf_buf *)m);
124#endif
125
126	KASSERT(curthread->td_pinned > 0 || (flags & SFB_CPUPRIVATE) == 0,
127	    ("sf_buf_alloc(SFB_CPUPRIVATE): curthread not pinned"));
128	hash_list = &sf_buf_active[SF_BUF_HASH(m)];
129	mtx_lock(&sf_buf_lock);
130	LIST_FOREACH(sf, hash_list, list_entry) {
131		if (sf->m == m) {
132			sf->ref_count++;
133			if (sf->ref_count == 1) {
134				TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry);
135				nsfbufsused++;
136				nsfbufspeak = imax(nsfbufspeak, nsfbufsused);
137			}
138#if defined(SMP) && defined(SFBUF_CPUSET)
139			sf_buf_shootdown(sf, flags);
140#endif
141			goto done;
142		}
143	}
144	while ((sf = TAILQ_FIRST(&sf_buf_freelist)) == NULL) {
145		if (flags & SFB_NOWAIT)
146			goto done;
147		sf_buf_alloc_want++;
148		SFSTAT_INC(sf_allocwait);
149		error = msleep(&sf_buf_freelist, &sf_buf_lock,
150		    (flags & SFB_CATCH) ? PCATCH | PVM : PVM, "sfbufa", 0);
151		sf_buf_alloc_want--;
152
153		/*
154		 * If we got a signal, don't risk going back to sleep.
155		 */
156		if (error)
157			goto done;
158	}
159	TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry);
160	if (sf->m != NULL)
161		LIST_REMOVE(sf, list_entry);
162	LIST_INSERT_HEAD(hash_list, sf, list_entry);
163	sf->ref_count = 1;
164	sf->m = m;
165	nsfbufsused++;
166	nsfbufspeak = imax(nsfbufspeak, nsfbufsused);
167	sf_buf_map(sf, flags);
168done:
169	mtx_unlock(&sf_buf_lock);
170	return (sf);
171}
172
173/*
174 * Remove a reference from the given sf_buf, adding it to the free
175 * list when its reference count reaches zero.  A freed sf_buf still,
176 * however, retains its virtual-to-physical mapping until it is
177 * recycled or reactivated by sf_buf_alloc(9).
178 */
179void
180sf_buf_free(struct sf_buf *sf)
181{
182
183#ifdef SFBUF_OPTIONAL_DIRECT_MAP
184	if (SFBUF_OPTIONAL_DIRECT_MAP)
185		return;
186#endif
187
188	mtx_lock(&sf_buf_lock);
189	sf->ref_count--;
190	if (sf->ref_count == 0) {
191		TAILQ_INSERT_TAIL(&sf_buf_freelist, sf, free_entry);
192		nsfbufsused--;
193		if (sf_buf_unmap(sf)) {
194			sf->m = NULL;
195			LIST_REMOVE(sf, list_entry);
196		}
197		if (sf_buf_alloc_want > 0)
198			wakeup(&sf_buf_freelist);
199	}
200	mtx_unlock(&sf_buf_lock);
201}
202
203void
204sf_buf_ref(struct sf_buf *sf)
205{
206
207#ifdef SFBUF_OPTIONAL_DIRECT_MAP
208	if (SFBUF_OPTIONAL_DIRECT_MAP)
209		return;
210#endif
211
212	mtx_lock(&sf_buf_lock);
213	KASSERT(sf->ref_count > 0, ("%s: sf %p not allocated", __func__, sf));
214	sf->ref_count++;
215	mtx_unlock(&sf_buf_lock);
216}
217
218#ifdef SFBUF_PROCESS_PAGE
219/*
220 * Run callback function on sf_buf that holds a certain page.
221 */
222boolean_t
223sf_buf_process_page(vm_page_t m, void (*cb)(struct sf_buf *))
224{
225	struct sf_head *hash_list;
226	struct sf_buf *sf;
227
228	hash_list = &sf_buf_active[SF_BUF_HASH(m)];
229	mtx_lock(&sf_buf_lock);
230	LIST_FOREACH(sf, hash_list, list_entry) {
231		if (sf->m == m) {
232			cb(sf);
233			mtx_unlock(&sf_buf_lock);
234			return (TRUE);
235		}
236	}
237	mtx_unlock(&sf_buf_lock);
238	return (FALSE);
239}
240#endif	/* SFBUF_PROCESS_PAGE */
241