uipc_cow.c revision 137372
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: head/sys/kern/uipc_cow.c 137372 2004-11-08 00:43:46Z alc $"); 40 41#include <sys/param.h> 42#include <sys/systm.h> 43#include <sys/kernel.h> 44#include <sys/proc.h> 45#include <sys/lock.h> 46#include <sys/mutex.h> 47#include <sys/mbuf.h> 48#include <sys/sf_buf.h> 49#include <sys/socketvar.h> 50#include <sys/uio.h> 51 52#include <vm/vm.h> 53#include <vm/vm_param.h> 54#include <vm/pmap.h> 55#include <vm/vm_map.h> 56#include <vm/vm_page.h> 57#include <vm/vm_object.h> 58 59 60struct netsend_cow_stats { 61 int attempted; 62 int fail_not_mapped; 63 int fail_sf_buf; 64 int success; 65 int iodone; 66}; 67 68static struct netsend_cow_stats socow_stats; 69 70static void socow_iodone(void *addr, void *args); 71 72static void 73socow_iodone(void *addr, void *args) 74{ 75 struct sf_buf *sf; 76 vm_page_t pp; 77 78 sf = args; 79 pp = sf_buf_page(sf); 80 sf_buf_free(sf); 81 /* remove COW mapping */ 82 vm_page_lock_queues(); 83 vm_page_cowclear(pp); 84 vm_page_unwire(pp, 0); 85 /* 86 * Check for the object going away on us. This can 87 * happen since we don't hold a reference to it. 88 * If so, we're responsible for freeing the page. 89 */ 90 if (pp->wire_count == 0 && pp->object == NULL) 91 vm_page_free(pp); 92 vm_page_unlock_queues(); 93 socow_stats.iodone++; 94} 95 96int 97socow_setup(struct mbuf *m0, struct uio *uio) 98{ 99 struct sf_buf *sf; 100 vm_page_t pp; 101 vm_paddr_t pa; 102 struct iovec *iov; 103 struct vmspace *vmspace; 104 struct vm_map *map; 105 vm_offset_t uva; 106 int s; 107 108 vmspace = curproc->p_vmspace; 109 map = &vmspace->vm_map; 110 uva = (vm_offset_t) uio->uio_iov->iov_base; 111 112 s = splvm(); 113 114 /* 115 * verify page is mapped & not already wired for i/o 116 */ 117 socow_stats.attempted++; 118 pa=pmap_extract(map->pmap, uva); 119 if(!pa) { 120 socow_stats.fail_not_mapped++; 121 splx(s); 122 return(0); 123 } 124 pp = PHYS_TO_VM_PAGE(pa); 125 126 /* 127 * set up COW 128 */ 129 vm_page_lock_queues(); 130 vm_page_cowsetup(pp); 131 132 /* 133 * wire the page for I/O 134 */ 135 vm_page_wire(pp); 136 vm_page_unlock_queues(); 137 138 /* 139 * Allocate an sf buf 140 */ 141 sf = sf_buf_alloc(pp, SFB_CATCH); 142 if (!sf) { 143 vm_page_lock_queues(); 144 vm_page_cowclear(pp); 145 vm_page_unwire(pp, 0); 146 /* 147 * Check for the object going away on us. This can 148 * happen since we don't hold a reference to it. 149 * If so, we're responsible for freeing the page. 150 */ 151 if (pp->wire_count == 0 && pp->object == NULL) 152 vm_page_free(pp); 153 vm_page_unlock_queues(); 154 socow_stats.fail_sf_buf++; 155 splx(s); 156 return(0); 157 } 158 /* 159 * attach to mbuf 160 */ 161 m0->m_data = (caddr_t)sf_buf_kva(sf); 162 m0->m_len = PAGE_SIZE; 163 MEXTADD(m0, sf_buf_kva(sf), PAGE_SIZE, socow_iodone, sf, M_RDONLY, 164 EXT_SFBUF); 165 socow_stats.success++; 166 167 iov = uio->uio_iov; 168 iov->iov_base = (char *)iov->iov_base + PAGE_SIZE; 169 iov->iov_len -= PAGE_SIZE; 170 uio->uio_resid -= PAGE_SIZE; 171 uio->uio_offset += PAGE_SIZE; 172 if (iov->iov_len == 0) { 173 uio->uio_iov++; 174 uio->uio_iovcnt--; 175 } 176 177 splx(s); 178 return(1); 179} 180