2 * Copyright (c) 2014 Gleb Smirnoff <glebius@FreeBSD.org>
3 * Copyright (c) 2003, 2005 Alan L. Cox <alc@cs.rice.edu>
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
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.
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
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
31 #include <sys/param.h>
32 #include <sys/kernel.h>
34 #include <sys/malloc.h>
35 #include <sys/mutex.h>
37 #include <sys/sf_buf.h>
39 #include <sys/sysctl.h>
42 #include <vm/vm_extern.h>
43 #include <vm/vm_page.h>
46 #define NSFBUFS (512 + maxusers * 16)
50 static int nsfbufspeak;
51 static int nsfbufsused;
53 SYSCTL_INT(_kern_ipc, OID_AUTO, nsfbufs, CTLFLAG_RDTUN, &nsfbufs, 0,
54 "Maximum number of sendfile(2) sf_bufs available");
55 SYSCTL_INT(_kern_ipc, OID_AUTO, nsfbufspeak, CTLFLAG_RD, &nsfbufspeak, 0,
56 "Number of sendfile(2) sf_bufs at peak usage");
57 SYSCTL_INT(_kern_ipc, OID_AUTO, nsfbufsused, CTLFLAG_RD, &nsfbufsused, 0,
58 "Number of sendfile(2) sf_bufs in use");
60 static void sf_buf_init(void *arg);
61 SYSINIT(sock_sf, SI_SUB_MBUF, SI_ORDER_ANY, sf_buf_init, NULL);
63 LIST_HEAD(sf_head, sf_buf);
66 * A hash table of active sendfile(2) buffers
68 static struct sf_head *sf_buf_active;
69 static u_long sf_buf_hashmask;
71 #define SF_BUF_HASH(m) (((m) - vm_page_array) & sf_buf_hashmask)
73 static TAILQ_HEAD(, sf_buf) sf_buf_freelist;
74 static u_int sf_buf_alloc_want;
77 * A lock used to synchronize access to the hash table and free list
79 static struct mtx sf_buf_lock;
82 * Allocate a pool of sf_bufs (sendfile(2) or "super-fast" if you prefer. :-))
85 sf_buf_init(void *arg)
87 struct sf_buf *sf_bufs;
91 #ifdef SFBUF_OPTIONAL_DIRECT_MAP
92 if (SFBUF_OPTIONAL_DIRECT_MAP)
97 TUNABLE_INT_FETCH("kern.ipc.nsfbufs", &nsfbufs);
99 sf_buf_active = hashinit(nsfbufs, M_TEMP, &sf_buf_hashmask);
100 TAILQ_INIT(&sf_buf_freelist);
101 sf_base = kva_alloc(nsfbufs * PAGE_SIZE);
102 sf_bufs = malloc(nsfbufs * sizeof(struct sf_buf), M_TEMP,
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);
108 sf_buf_alloc_want = 0;
109 mtx_init(&sf_buf_lock, "sf_buf", NULL, MTX_DEF);
113 * Get an sf_buf from the freelist. May block if none are available.
116 sf_buf_alloc(struct vm_page *m, int flags)
118 struct sf_head *hash_list;
122 #ifdef SFBUF_OPTIONAL_DIRECT_MAP
123 if (SFBUF_OPTIONAL_DIRECT_MAP)
124 return ((struct sf_buf *)m);
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) {
134 if (sf->ref_count == 1) {
135 TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry);
137 nsfbufspeak = imax(nsfbufspeak, nsfbufsused);
139 #if defined(SMP) && defined(SFBUF_CPUSET)
140 sf_buf_shootdown(sf, flags);
145 while ((sf = TAILQ_FIRST(&sf_buf_freelist)) == NULL) {
146 if (flags & SFB_NOWAIT)
149 SFSTAT_INC(sf_allocwait);
150 error = msleep(&sf_buf_freelist, &sf_buf_lock,
151 (flags & SFB_CATCH) ? PCATCH | PVM : PVM, "sfbufa", 0);
155 * If we got a signal, don't risk going back to sleep.
160 TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry);
162 LIST_REMOVE(sf, list_entry);
163 LIST_INSERT_HEAD(hash_list, sf, list_entry);
167 nsfbufspeak = imax(nsfbufspeak, nsfbufsused);
168 sf_buf_map(sf, flags);
170 mtx_unlock(&sf_buf_lock);
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).
181 sf_buf_free(struct sf_buf *sf)
184 #ifdef SFBUF_OPTIONAL_DIRECT_MAP
185 if (SFBUF_OPTIONAL_DIRECT_MAP)
189 mtx_lock(&sf_buf_lock);
191 if (sf->ref_count == 0) {
192 TAILQ_INSERT_TAIL(&sf_buf_freelist, sf, free_entry);
194 if (sf_buf_unmap(sf)) {
196 LIST_REMOVE(sf, list_entry);
198 if (sf_buf_alloc_want > 0)
199 wakeup(&sf_buf_freelist);
201 mtx_unlock(&sf_buf_lock);
205 sf_buf_ref(struct sf_buf *sf)
208 #ifdef SFBUF_OPTIONAL_DIRECT_MAP
209 if (SFBUF_OPTIONAL_DIRECT_MAP)
213 mtx_lock(&sf_buf_lock);
214 KASSERT(sf->ref_count > 0, ("%s: sf %p not allocated", __func__, sf));
216 mtx_unlock(&sf_buf_lock);
219 #ifdef SFBUF_PROCESS_PAGE
221 * Run callback function on sf_buf that holds a certain page.
224 sf_buf_process_page(vm_page_t m, void (*cb)(struct sf_buf *))
226 struct sf_head *hash_list;
229 hash_list = &sf_buf_active[SF_BUF_HASH(m)];
230 mtx_lock(&sf_buf_lock);
231 LIST_FOREACH(sf, hash_list, list_entry) {
234 mtx_unlock(&sf_buf_lock);
238 mtx_unlock(&sf_buf_lock);
241 #endif /* SFBUF_PROCESS_PAGE */