rx: Process error queue after noticing errors

[openafs.git] / src / rx / LINUX / rx_knet.c

diff --git a/src/rx/LINUX/rx_knet.c b/src/rx/LINUX/rx_knet.c
index d93d2d5..48a493d 100644 (file)
--- a/src/rx/LINUX/rx_knet.c
+++ b/src/rx/LINUX/rx_knet.c
@@ -18,9 +18,15 @@
 
 #include <linux/version.h>
 #include "rx/rx_kcommon.h"
-#include "h/smp_lock.h"
+#include "rx.h"
+#include "rx_atomic.h"
+#include "rx_globals.h"
+#include "rx_stats.h"
+#include "rx_peer.h"
+#include "rx_packet.h"
+#include "rx_internal.h"
 #include <asm/uaccess.h>
-#ifdef ADAPT_PMTU
+#ifdef AFS_RXERRQ_ENV
 #include <linux/errqueue.h>
 #include <linux/icmp.h>
 #endif
@@ -36,9 +42,8 @@ rxk_NewSocketHost(afs_uint32 ahost, short aport)
     struct socket *sockp;
     struct sockaddr_in myaddr;
     int code;
-#ifdef ADAPT_PMTU
+#ifdef AFS_ADAPT_PMTU
     int pmtu = IP_PMTUDISC_WANT;
-    int do_recverr = 1;
 #else
     int pmtu = IP_PMTUDISC_DONT;
 #endif
@@ -68,9 +73,12 @@ rxk_NewSocketHost(afs_uint32 ahost, short aport)
 
     kernel_setsockopt(sockp, SOL_IP, IP_MTU_DISCOVER, (char *)&pmtu,
                      sizeof(pmtu));
-#ifdef ADAPT_PMTU
-    kernel_setsockopt(sockp, SOL_IP, IP_RECVERR, (char *)&do_recverr,
-                      sizeof(do_recverr));
+#ifdef AFS_RXERRQ_ENV
+    {
+       int recverr = 1;
+       kernel_setsockopt(sockp, SOL_IP, IP_RECVERR, (char *)&recverr,
+                         sizeof(recverr));
+    }
 #endif
     return (osi_socket *)sockp;
 }
@@ -89,40 +97,48 @@ rxk_FreeSocket(struct socket *asocket)
     return 0;
 }
 
-#ifdef ADAPT_PMTU
-void
-handle_socket_error(osi_socket so)
+#ifdef AFS_RXERRQ_ENV
+int
+osi_HandleSocketError(osi_socket so)
 {
+    int ret = 0;
     struct msghdr msg;
     struct cmsghdr *cmsg;
     struct sock_extended_err *err;
     struct sockaddr_in addr;
     struct sockaddr *offender;
-    char *controlmsgbuf;
+    char *controlmsgbuf = NULL;
     int code;
     struct socket *sop = (struct socket *)so;
 
-    if (!(controlmsgbuf=rxi_Alloc(256)))
-       return;
+    if (!(controlmsgbuf = rxi_Alloc(256)))
+       goto out;
     msg.msg_name = &addr;
     msg.msg_namelen = sizeof(addr);
     msg.msg_control = controlmsgbuf;
     msg.msg_controllen = 256;
     msg.msg_flags = 0;
 
-    code = kernel_recvmsg(sop, &msg, NULL, 0, 256,
+    code = kernel_recvmsg(sop, &msg, NULL, 0, 0,
                          MSG_ERRQUEUE|MSG_DONTWAIT|MSG_TRUNC);
 
     if (code < 0 || !(msg.msg_flags & MSG_ERRQUEUE))
        goto out;
 
-    for (cmsg = CMSG_FIRSTHDR(&msg); cmsg; cmsg = CMSG_NXTHDR(&msg, cmsg)) {
-       if (CMSG_OK(&msg, cmsg) && cmsg->cmsg_level == SOL_IP &&
-           cmsg->cmsg_type == IP_RECVERR)
+    /* kernel_recvmsg changes msg_control to point at the _end_ of the buffer,
+     * and msg_controllen is set to the number of bytes remaining */
+    msg.msg_controllen = ((char*)msg.msg_control - (char*)controlmsgbuf);
+    msg.msg_control = controlmsgbuf;
+
+    for (cmsg = CMSG_FIRSTHDR(&msg); cmsg && CMSG_OK(&msg, cmsg);
+         cmsg = CMSG_NXTHDR(&msg, cmsg)) {
+       if (cmsg->cmsg_level == SOL_IP && cmsg->cmsg_type == IP_RECVERR)
            break;
     }
     if (!cmsg)
        goto out;
+
+    ret = 1;
     err = CMSG_DATA(cmsg);
     offender = SO_EE_OFFENDER(err);
     
@@ -131,17 +147,13 @@ handle_socket_error(osi_socket so)
 
     memcpy(&addr, offender, sizeof(addr));
 
-    if (err->ee_origin == SO_EE_ORIGIN_ICMP &&
-       err->ee_type == ICMP_DEST_UNREACH &&
-       err->ee_code == ICMP_FRAG_NEEDED) {
-       rxi_SetPeerMtu(ntohl(addr.sin_addr.s_addr), ntohs(addr.sin_port),
-                      err->ee_info);
-    }
-    /* other DEST_UNREACH's and TIME_EXCEEDED should be dealt with too */
+    rxi_ProcessNetError(err, addr.sin_addr.s_addr, addr.sin_port);
 
-out:
-    rxi_Free(controlmsgbuf, 256);
-    return;
+ out:
+    if (controlmsgbuf) {
+       rxi_Free(controlmsgbuf, 256);
+    }
+    return ret;
 }
 #endif
 
@@ -157,19 +169,7 @@ osi_NetSend(osi_socket sop, struct sockaddr_in *to, struct iovec *iovec,
 {
     struct msghdr msg;
     int code;
-#ifdef ADAPT_PMTU
-    int sockerr;
-    size_t esize;
-
-    while (1) {
-       sockerr=0;
-       esize = sizeof(sockerr);
-       kernel_getsockopt(sop, SOL_SOCKET, SO_ERROR, (char *)&sockerr, &esize);
-       if (sockerr == 0)
-          break;
-       handle_socket_error(sop);
-    }
-#endif
+
 
     msg.msg_name = to;
     msg.msg_namelen = sizeof(*to);
@@ -178,6 +178,14 @@ osi_NetSend(osi_socket sop, struct sockaddr_in *to, struct iovec *iovec,
     msg.msg_flags = 0;
 
     code = kernel_sendmsg(sop, &msg, (struct kvec *) iovec, iovcnt, size);
+
+#ifdef AFS_RXERRQ_ENV
+    if (code < 0) {
+       while (osi_HandleSocketError(sop))
+           ;
+    }
+#endif
+
     return (code < 0) ? code : 0;
 }
 
@@ -209,26 +217,13 @@ osi_NetReceive(osi_socket so, struct sockaddr_in *from, struct iovec *iov,
 {
     struct msghdr msg;
     int code;
-#ifdef ADAPT_PMTU
-    int sockerr;
-    size_t esize;
-#endif
     struct iovec tmpvec[RX_MAXWVECS + 2];
     struct socket *sop = (struct socket *)so;
 
     if (iovcnt > RX_MAXWVECS + 2) {
        osi_Panic("Too many (%d) iovecs passed to osi_NetReceive\n", iovcnt);
     }
-#ifdef ADAPT_PMTU
-    while (1) {
-       sockerr=0;
-       esize = sizeof(sockerr);
-       kernel_getsockopt(sop, SOL_SOCKET, SO_ERROR, (char *)&sockerr, &esize);
-       if (sockerr == 0)
-          break;
-       handle_socket_error(so);
-    }
-#endif
+
     memcpy(tmpvec, iov, iovcnt * sizeof(struct iovec));
     msg.msg_name = from;
     msg.msg_iov = tmpvec;
@@ -240,29 +235,7 @@ osi_NetReceive(osi_socket so, struct sockaddr_in *from, struct iovec *iov,
     code = kernel_recvmsg(sop, &msg, (struct kvec *)tmpvec, iovcnt,
                          *lengthp, 0);
     if (code < 0) {
-#ifdef CONFIG_PM
-       if (
-# ifdef PF_FREEZE
-           current->flags & PF_FREEZE
-# else
-#  if defined(STRUCT_TASK_STRUCT_HAS_TODO)
-           !current->todo
-#  else
-#   if defined(STRUCT_TASK_STRUCT_HAS_THREAD_INFO)
-            test_ti_thread_flag(current->thread_info, TIF_FREEZE)
-#   else
-            test_ti_thread_flag(task_thread_info(current), TIF_FREEZE)
-#   endif
-#  endif
-# endif
-           )
-# ifdef LINUX_REFRIGERATOR_TAKES_PF_FREEZE
-           refrigerator(PF_FREEZE);
-# else
-           refrigerator();
-# endif
-           set_current_state(TASK_INTERRUPTIBLE);
-#endif
+       afs_try_to_freeze();
 
        /* Clear the error before using the socket again.
         * Oh joy, Linux has hidden header files as well. It appears we can
@@ -271,6 +244,11 @@ osi_NetReceive(osi_socket so, struct sockaddr_in *from, struct iovec *iov,
        flush_signals(current); /* We don't want no stinkin' signals. */
        rxk_lastSocketError = code;
        rxk_nSocketErrors++;
+
+#ifdef AFS_RXERRQ_ENV
+       while (osi_HandleSocketError(so))
+           ;
+#endif
     } else {
        *lengthp = code;
        code = 0;