subr_sfbuf.c revision 269807 
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: head/sys/kern/subr_sfbuf.c 269807 2014-08-11 12:59:55Z 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_NOWAIT | M_ZERO);
103	KASSERT(sf_bufs, ("%s: malloc failure", __func__));
104	for (i = 0; i < nsfbufs; i++) {
105		sf_bufs[i].kva = sf_base + i * PAGE_SIZE;
106		TAILQ_INSERT_TAIL(&sf_buf_freelist, &sf_bufs[i], free_entry);
107	}
108	sf_buf_alloc_want = 0;
109	mtx_init(&sf_buf_lock, "sf_buf", NULL, MTX_DEF);
110}
111
112/*
113 * Get an sf_buf from the freelist.  May block if none are available.
114 */
115struct sf_buf *
116sf_buf_alloc(struct vm_page *m, int flags)
117{
118	struct sf_head *hash_list;
119	struct sf_buf *sf;
120	int error;
121
122#ifdef SFBUF_OPTIONAL_DIRECT_MAP
123	if (SFBUF_OPTIONAL_DIRECT_MAP)
124		return ((struct sf_buf *)m);
125#endif
126
127	KASSERT(curthread->td_pinned > 0 || (flags & SFB_CPUPRIVATE) == 0,
128	    ("sf_buf_alloc(SFB_CPUPRIVATE): curthread not pinned"));
129	hash_list = &sf_buf_active[SF_BUF_HASH(m)];
130	mtx_lock(&sf_buf_lock);
131	LIST_FOREACH(sf, hash_list, list_entry) {
132		if (sf->m == m) {
133			sf->ref_count++;
134			if (sf->ref_count == 1) {
135				TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry);
136				nsfbufsused++;
137				nsfbufspeak = imax(nsfbufspeak, nsfbufsused);
138			}
139#if defined(SMP) && defined(SFBUF_CPUSET)
140			sf_buf_shootdown(sf, flags);
141#endif
142			goto done;
143		}
144	}
145	while ((sf = TAILQ_FIRST(&sf_buf_freelist)) == NULL) {
146		if (flags & SFB_NOWAIT)
147			goto done;
148		sf_buf_alloc_want++;
149		SFSTAT_INC(sf_allocwait);
150		error = msleep(&sf_buf_freelist, &sf_buf_lock,
151		    (flags & SFB_CATCH) ? PCATCH | PVM : PVM, "sfbufa", 0);
152		sf_buf_alloc_want--;
153
154		/*
155		 * If we got a signal, don't risk going back to sleep.
156		 */
157		if (error)
158			goto done;
159	}
160	TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry);
161	if (sf->m != NULL)
162		LIST_REMOVE(sf, list_entry);
163	LIST_INSERT_HEAD(hash_list, sf, list_entry);
164	sf->ref_count = 1;
165	sf->m = m;
166	nsfbufsused++;
167	nsfbufspeak = imax(nsfbufspeak, nsfbufsused);
168	sf_buf_map(sf, flags);
169done:
170	mtx_unlock(&sf_buf_lock);
171	return (sf);
172}
173
174/*
175 * Remove a reference from the given sf_buf, adding it to the free
176 * list when its reference count reaches zero.  A freed sf_buf still,
177 * however, retains its virtual-to-physical mapping until it is
178 * recycled or reactivated by sf_buf_alloc(9).
179 */
180void
181sf_buf_free(struct sf_buf *sf)
182{
183
184#ifdef SFBUF_OPTIONAL_DIRECT_MAP
185	if (SFBUF_OPTIONAL_DIRECT_MAP)
186		return;
187#endif
188
189	mtx_lock(&sf_buf_lock);
190	sf->ref_count--;
191	if (sf->ref_count == 0) {
192		TAILQ_INSERT_TAIL(&sf_buf_freelist, sf, free_entry);
193		nsfbufsused--;
194		if (sf_buf_unmap(sf)) {
195			sf->m = NULL;
196			LIST_REMOVE(sf, list_entry);
197		}
198		if (sf_buf_alloc_want > 0)
199			wakeup(&sf_buf_freelist);
200	}
201	mtx_unlock(&sf_buf_lock);
202}
203
204void
205sf_buf_ref(struct sf_buf *sf)
206{
207
208#ifdef SFBUF_OPTIONAL_DIRECT_MAP
209	if (SFBUF_OPTIONAL_DIRECT_MAP)
210		return;
211#endif
212
213	KASSERT(sf->ref_count > 0, ("%s: sf %p not allocated", __func__, sf));
214
215	mtx_lock(&sf_buf_lock);
216	sf->ref_count++;
217	mtx_unlock(&sf_buf_lock);
218}
219
220#ifdef SFBUF_PROCESS_PAGE
221/*
222 * Run callback function on sf_buf that holds a certain page.
223 */
224boolean_t
225sf_buf_process_page(vm_page_t m, void (*cb)(struct sf_buf *))
226{
227	struct sf_head *hash_list;
228	struct sf_buf *sf;
229
230	hash_list = &sf_buf_active[SF_BUF_HASH(m)];
231	mtx_lock(&sf_buf_lock);
232	LIST_FOREACH(sf, hash_list, list_entry) {
233		if (sf->m == m) {
234			cb(sf);
235			mtx_unlock(&sf_buf_lock);
236			return (TRUE);
237		}
238	}
239	mtx_unlock(&sf_buf_lock);
240	return (FALSE);
241}
242#endif	/* SFBUF_PROCESS_PAGE */
243