src/rx/LINUX/rx_knet.c

   1 /*
   2  * Copyright 2000, International Business Machines Corporation and others.
   3  * All Rights Reserved.
   4  *
   5  * This software has been released under the terms of the IBM Public
   6  * License.  For details, see the LICENSE file in the top-level source
   7  * directory or online at http://www.openafs.org/dl/license10.html
   8  */
   9
  10 /*
  11  * rx_knet.c - RX kernel send, receive and timer routines.
  12  *
  13  * Linux implementation.
  14  */
  15 #include <afsconfig.h>
  16 #include "afs/param.h"
  17
  18
  19 #include <linux/version.h>
  20 #include "rx/rx_kcommon.h"
  21 #include "rx.h"
  22 #include "rx_atomic.h"
  23 #include "rx_globals.h"
  24 #include "rx_stats.h"
  25 #include "rx_peer.h"
  26 #include "rx_packet.h"
  27 #include "rx_internal.h"
  28 #if defined(HAVE_LINUX_UACCESS_H)
  29 #include <linux/uaccess.h>
  30 #else
  31 #include <asm/uaccess.h>
  32 #endif
  33 #ifdef AFS_RXERRQ_ENV
  34 #include <linux/errqueue.h>
  35 #include <linux/icmp.h>
  36 #endif
  37
  38 #include "osi_compat.h"
  39
  40 /* rxk_NewSocket
  41  * open and bind RX socket
  42  */
  43 osi_socket *
  44 rxk_NewSocketHost(afs_uint32 ahost, short aport)
  45 {
  46     struct socket *sockp;
  47     struct sockaddr_in myaddr;
  48     int code;
  49 #ifdef AFS_ADAPT_PMTU
  50     int pmtu = IP_PMTUDISC_WANT;
  51 #else
  52     int pmtu = IP_PMTUDISC_DONT;
  53 #endif
  54
  55 #ifdef HAVE_LINUX_SOCK_CREATE_KERN_NS
  56     code = sock_create_kern(&init_net, AF_INET, SOCK_DGRAM, IPPROTO_UDP, &sockp);
  57 #elif defined(HAVE_LINUX_SOCK_CREATE_KERN)
  58     code = sock_create_kern(AF_INET, SOCK_DGRAM, IPPROTO_UDP, &sockp);
  59 #elif defined(LINUX_KERNEL_SOCK_CREATE_V)
  60     code = sock_create(AF_INET, SOCK_DGRAM, IPPROTO_UDP, &sockp, 0);
  61 #else
  62     code = sock_create(AF_INET, SOCK_DGRAM, IPPROTO_UDP, &sockp);
  63 #endif
  64     if (code < 0)
  65         return NULL;
  66
  67     /* Bind socket */
  68     myaddr.sin_family = AF_INET;
  69     myaddr.sin_addr.s_addr = ahost;
  70     myaddr.sin_port = aport;
  71     code =
  72         sockp->ops->bind(sockp, (struct sockaddr *)&myaddr, sizeof(myaddr));
  73
  74     if (code < 0) {
  75         printk("sock_release(rx_socket) FIXME\n");
  76         return NULL;
  77     }
  78
  79     kernel_setsockopt(sockp, SOL_IP, IP_MTU_DISCOVER, (char *)&pmtu,
  80                       sizeof(pmtu));
  81 #ifdef AFS_RXERRQ_ENV
  82     {
  83         int recverr = 1;
  84         kernel_setsockopt(sockp, SOL_IP, IP_RECVERR, (char *)&recverr,
  85                           sizeof(recverr));
  86     }
  87 #endif
  88     return (osi_socket *)sockp;
  89 }
  90
  91 osi_socket *
  92 rxk_NewSocket(short aport)
  93 {
  94     return rxk_NewSocketHost(htonl(INADDR_ANY), aport);
  95 }
  96
  97 /* free socket allocated by osi_NetSocket */
  98 int
  99 rxk_FreeSocket(struct socket *asocket)
 100 {
 101     AFS_STATCNT(osi_FreeSocket);
 102     return 0;
 103 }
 104
 105 #ifdef AFS_RXERRQ_ENV
 106 static int
 107 osi_HandleSocketError(osi_socket so, char *cmsgbuf, size_t cmsgbuf_len)
 108 {
 109     struct msghdr msg;
 110     struct cmsghdr *cmsg;
 111     struct sock_extended_err *err;
 112     struct sockaddr_in addr;
 113     int code;
 114     struct socket *sop = (struct socket *)so;
 115
 116     msg.msg_name = &addr;
 117     msg.msg_namelen = sizeof(addr);
 118     msg.msg_control = cmsgbuf;
 119     msg.msg_controllen = cmsgbuf_len;
 120     msg.msg_flags = 0;
 121
 122     code = kernel_recvmsg(sop, &msg, NULL, 0, 0,
 123                           MSG_ERRQUEUE|MSG_DONTWAIT|MSG_TRUNC);
 124
 125     if (code < 0 || !(msg.msg_flags & MSG_ERRQUEUE))
 126         return 0;
 127
 128     /* kernel_recvmsg changes msg_control to point at the _end_ of the buffer,
 129      * and msg_controllen is set to the number of bytes remaining */
 130     msg.msg_controllen = ((char*)msg.msg_control - (char*)cmsgbuf);
 131     msg.msg_control = cmsgbuf;
 132
 133     for (cmsg = CMSG_FIRSTHDR(&msg); cmsg && CMSG_OK(&msg, cmsg);
 134          cmsg = CMSG_NXTHDR(&msg, cmsg)) {
 135
 136         if (cmsg->cmsg_level != SOL_IP || cmsg->cmsg_type != IP_RECVERR) {
 137             continue;
 138         }
 139
 140         err = CMSG_DATA(cmsg);
 141         rxi_ProcessNetError(err, addr.sin_addr.s_addr, addr.sin_port);
 142     }
 143
 144     return 1;
 145 }
 146 #endif
 147
 148 static void
 149 do_handlesocketerror(osi_socket so)
 150 {
 151 #ifdef AFS_RXERRQ_ENV
 152     char *cmsgbuf;
 153     size_t cmsgbuf_len;
 154
 155     cmsgbuf_len = 256;
 156     cmsgbuf = rxi_Alloc(cmsgbuf_len);
 157     if (!cmsgbuf) {
 158         return;
 159     }
 160
 161     while (osi_HandleSocketError(so, cmsgbuf, cmsgbuf_len))
 162         ;
 163
 164     rxi_Free(cmsgbuf, cmsgbuf_len);
 165 #endif
 166 }
 167
 168 /* osi_NetSend
 169  *
 170  * Return codes:
 171  * 0 = success
 172  * non-zero = failure
 173  */
 174 int
 175 osi_NetSend(osi_socket sop, struct sockaddr_in *to, struct iovec *iovec,
 176             int iovcnt, afs_int32 size, int istack)
 177 {
 178     struct msghdr msg;
 179     int code;
 180
 181
 182     msg.msg_name = to;
 183     msg.msg_namelen = sizeof(*to);
 184     msg.msg_control = NULL;
 185     msg.msg_controllen = 0;
 186     msg.msg_flags = 0;
 187
 188     code = kernel_sendmsg(sop, &msg, (struct kvec *) iovec, iovcnt, size);
 189
 190     if (code < 0) {
 191         do_handlesocketerror(sop);
 192     }
 193
 194     return (code < 0) ? code : 0;
 195 }
 196
 197
 198 /* osi_NetReceive
 199  * OS dependent part of kernel RX listener thread.
 200  *
 201  * Arguments:
 202  *      so      socket to receive on, typically rx_socket
 203  *      from    pointer to a sockaddr_in.
 204  *      iov     array of iovecs to fill in.
 205  *      iovcnt  how many iovecs there are.
 206  *      lengthp IN/OUT in: total space available in iovecs. out: size of read.
 207  *
 208  * Return
 209  * 0 if successful
 210  * error code (such as EINTER) if not
 211  *
 212  * Environment
 213  *      Note that the maximum number of iovecs is 2 + RX_MAXWVECS. This is
 214  *      so we have a little space to look for packets larger than
 215  *      rx_maxReceiveSize.
 216  */
 217 int rxk_lastSocketError;
 218 int rxk_nSocketErrors;
 219 int
 220 osi_NetReceive(osi_socket so, struct sockaddr_in *from, struct iovec *iov,
 221                int iovcnt, int *lengthp)
 222 {
 223     struct msghdr msg;
 224     int code;
 225     struct iovec tmpvec[RX_MAXWVECS + 2];
 226     struct socket *sop = (struct socket *)so;
 227
 228     if (iovcnt > RX_MAXWVECS + 2) {
 229         osi_Panic("Too many (%d) iovecs passed to osi_NetReceive\n", iovcnt);
 230     }
 231
 232     memcpy(tmpvec, iov, iovcnt * sizeof(struct iovec));
 233     msg.msg_name = from;
 234 #if defined(STRUCT_MSGHDR_HAS_MSG_ITER)
 235     msg.msg_iter.iov = tmpvec;
 236     msg.msg_iter.nr_segs = iovcnt;
 237 #else
 238     msg.msg_iov = tmpvec;
 239     msg.msg_iovlen = iovcnt;
 240 #endif
 241     msg.msg_control = NULL;
 242     msg.msg_controllen = 0;
 243     msg.msg_flags = 0;
 244
 245     code = kernel_recvmsg(sop, &msg, (struct kvec *)tmpvec, iovcnt,
 246                           *lengthp, 0);
 247     if (code < 0) {
 248         afs_try_to_freeze();
 249
 250         /* Clear the error before using the socket again.
 251          * Oh joy, Linux has hidden header files as well. It appears we can
 252          * simply call again and have it clear itself via sock_error().
 253          */
 254         flush_signals(current); /* We don't want no stinkin' signals. */
 255         rxk_lastSocketError = code;
 256         rxk_nSocketErrors++;
 257
 258         do_handlesocketerror(so);
 259     } else {
 260         *lengthp = code;
 261         code = 0;
 262     }
 263
 264     return code;
 265 }
 266
 267 void
 268 osi_StopListener(void)
 269 {
 270     extern struct task_struct *rxk_ListenerTask;
 271
 272     while (rxk_ListenerTask) {
 273         if (rxk_ListenerTask) {
 274             flush_signals(rxk_ListenerTask);
 275             send_sig(SIGKILL, rxk_ListenerTask, 1);
 276         }
 277         if (!rxk_ListenerTask)
 278             break;
 279         afs_osi_Sleep(&rxk_ListenerTask);
 280     }
 281     sock_release(rx_socket);
 282     rx_socket = NULL;
 283 }
 284