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