sys/kern/uipc_cow.c

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