1 /*--
2 * Copyright (c) 1997, Duke University
3 * All rights reserved.
4 *
5 * Author:
6 * Andrew Gallatin <gallatin@cs.duke.edu>
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. The name of Duke University may not be used to endorse or promote
17 * products derived from this software without specific prior written
18 * permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY DUKE UNIVERSITY ``AS IS'' AND ANY
21 * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL DUKE UNIVERSITY BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITSOR BUSINESS
27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER
28 * IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
29 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
30 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
31 */
32
33 /*
34 * This is a set of routines for enabling and disabling copy on write
35 * protection for data written into sockets.
36 */
37
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD: stable/9/sys/kern/uipc_cow.c 219028 2011-02-25 10:11:01Z netchild $");
40
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/sysctl.h>
44 #include <sys/kernel.h>
45 #include <sys/proc.h>
46 #include <sys/lock.h>
47 #include <sys/mutex.h>
48 #include <sys/mbuf.h>
49 #include <sys/sf_buf.h>
50 #include <sys/socketvar.h>
51 #include <sys/uio.h>
52
53 #include <vm/vm.h>
54 #include <vm/vm_extern.h>
55 #include <vm/vm_param.h>
56 #include <vm/pmap.h>
57 #include <vm/vm_map.h>
58 #include <vm/vm_page.h>
59 #include <vm/vm_object.h>
60
61 FEATURE(zero_copy_sockets, "Zero copy sockets support");
62
63 struct netsend_cow_stats {
64 int attempted;
65 int fail_not_mapped;
66 int fail_sf_buf;
67 int success;
68 int iodone;
69 };
70
71 static struct netsend_cow_stats socow_stats;
72
73 static void socow_iodone(void *addr, void *args);
74
75 static void
socow_iodone(void * addr,void * args)76 socow_iodone(void *addr, void *args)
77 {
78 struct sf_buf *sf;
79 vm_page_t pp;
80
81 sf = args;
82 pp = sf_buf_page(sf);
83 sf_buf_free(sf);
84 /* remove COW mapping */
85 vm_page_lock(pp);
86 vm_page_cowclear(pp);
87 vm_page_unwire(pp, 0);
88 /*
89 * Check for the object going away on us. This can
90 * happen since we don't hold a reference to it.
91 * If so, we're responsible for freeing the page.
92 */
93 if (pp->wire_count == 0 && pp->object == NULL)
94 vm_page_free(pp);
95 vm_page_unlock(pp);
96 socow_stats.iodone++;
97 }
98
99 int
socow_setup(struct mbuf * m0,struct uio * uio)100 socow_setup(struct mbuf *m0, struct uio *uio)
101 {
102 struct sf_buf *sf;
103 vm_page_t pp;
104 struct iovec *iov;
105 struct vmspace *vmspace;
106 struct vm_map *map;
107 vm_offset_t offset, uva;
108 vm_size_t len;
109
110 socow_stats.attempted++;
111 vmspace = curproc->p_vmspace;
112 map = &vmspace->vm_map;
113 uva = (vm_offset_t) uio->uio_iov->iov_base;
114 offset = uva & PAGE_MASK;
115 len = PAGE_SIZE - offset;
116
117 /*
118 * Verify that access to the given address is allowed from user-space.
119 */
120 if (vm_fault_quick_hold_pages(map, uva, len, VM_PROT_READ, &pp, 1) <
121 0) {
122 socow_stats.fail_not_mapped++;
123 return(0);
124 }
125
126 /*
127 * set up COW
128 */
129 vm_page_lock(pp);
130 if (vm_page_cowsetup(pp) != 0) {
131 vm_page_unhold(pp);
132 vm_page_unlock(pp);
133 return (0);
134 }
135
136 /*
137 * wire the page for I/O
138 */
139 vm_page_wire(pp);
140 vm_page_unhold(pp);
141 vm_page_unlock(pp);
142 /*
143 * Allocate an sf buf
144 */
145 sf = sf_buf_alloc(pp, SFB_CATCH);
146 if (sf == NULL) {
147 vm_page_lock(pp);
148 vm_page_cowclear(pp);
149 vm_page_unwire(pp, 0);
150 /*
151 * Check for the object going away on us. This can
152 * happen since we don't hold a reference to it.
153 * If so, we're responsible for freeing the page.
154 */
155 if (pp->wire_count == 0 && pp->object == NULL)
156 vm_page_free(pp);
157 vm_page_unlock(pp);
158 socow_stats.fail_sf_buf++;
159 return(0);
160 }
161 /*
162 * attach to mbuf
163 */
164 MEXTADD(m0, sf_buf_kva(sf), PAGE_SIZE, socow_iodone,
165 (void*)sf_buf_kva(sf), sf, M_RDONLY, EXT_SFBUF);
166 m0->m_len = len;
167 m0->m_data = (caddr_t)sf_buf_kva(sf) + offset;
168 socow_stats.success++;
169
170 iov = uio->uio_iov;
171 iov->iov_base = (char *)iov->iov_base + m0->m_len;
172 iov->iov_len -= m0->m_len;
173 uio->uio_resid -= m0->m_len;
174 uio->uio_offset += m0->m_len;
175 if (iov->iov_len == 0) {
176 uio->uio_iov++;
177 uio->uio_iovcnt--;
178 }
179
180 return(m0->m_len);
181 }
182