Extend the parsing capability of the oob tcp module's if_include and if_exclude options to support subnet+mask notation, and to handle virtual IP addresses (it was previously having problems distinguishing between "eth1" and "eth1.3").

This commit was SVN r24747.

Author: Ralph Castain

opal/util/help-opal-util.txt

@@ -35,3 +35,18 @@ of the form <ipaddress>/<mask>.  For example:
 
 All malformed entries will be ignored; Open MPI will attempt to continue
 your job.  The first detected malformed entry was %s.
+#
+[invalid-net-mask]
+We were unable to parse the provided network interface:
+
+  Interface: %s
+
+The interface must be one of the following forms:
+
+  123.456.789.123
+  123.456/16
+  123.456.789
+
+The system can parse any one of these, and will find an interface
+that matches within the provided scope. Please revise your input
+and try again.

opal/util/if.c

@@ -68,11 +68,13 @@
 #ifdef HAVE_IFADDRS_H
 #include <ifaddrs.h>
 #endif
+#include <ctype.h>
 
 #include "opal/class/opal_list.h"
 #include "opal/util/if.h"
 #include "opal/util/output.h"
 #include "opal/util/argv.h"
+#include "opal/util/show_help.h"
 #include "opal/constants.h"
 
 #include "opal/mca/if/base/base.h"
@@ -601,6 +603,57 @@ bool opal_ifisloopback(int if_index)
     return false;
 }
 
+/* Determine if an interface matches any entry in the given list, taking
+ * into account that the list entries could be given as named interfaces,
+ * IP addrs, or subnet+mask
+ */
+bool opal_ifmatches(int idx, char **nets)
+{
+    bool named_if;
+    int i;
+    size_t j;
+    int index;
+    struct sockaddr_in inaddr;
+    uint32_t addr, netaddr, netmask;
+
+    /* get the address info for the given network in case we need it */
+    if (OPAL_SUCCESS != opal_ifindextoaddr(idx, (struct sockaddr*)&inaddr, sizeof(inaddr))) {
+        return false;
+    }
+    addr = ntohl(inaddr.sin_addr.s_addr);
+
+    for (i=0; NULL != nets[i]; i++) {
+        /* if the specified interface contains letters in it, then it
+         * was given as an interface name and not an IP tuple
+         */
+        named_if = false;
+        for (j=0; j < strlen(nets[i]); j++) {
+            if (isalpha(nets[i][j]) && '.' != nets[i][j]) {
+                named_if = true;
+                break;
+            }
+        }
+        if (named_if) {
+            if (0 > (index = opal_ifnametoindex(nets[i]))) {
+                continue;
+            }
+            if (index == idx) {
+                return true;
+            }
+        } else {
+            if (OPAL_SUCCESS != opal_iftupletoaddr(nets[i], &netaddr, &netmask)) {
+                opal_show_help("help-opal-util.txt", "invalid-net-mask", true, nets[i]);
+                continue;
+            }
+            if (netaddr == (addr & netmask)) {
+                return true;
+            }
+        }
+    }
+    /* get here if not found */
+    return false;
+}
+
 
 #else /* HAVE_STRUCT_SOCKADDR_IN */
 
@@ -693,5 +746,10 @@ opal_iftupletoaddr(char *inaddr, uint32_t *net, uint32_t *mask)
     return 0;
 }
 
+bool opal_ifispresent(char *if)
+{
+    return false;
+}
+
 #endif /* HAVE_STRUCT_SOCKADDR_IN */
 

opal/util/if.h

@@ -188,6 +188,10 @@ OPAL_DECLSPEC int opal_iftupletoaddr(const char *addr, uint32_t *net, uint32_t *
  */
 OPAL_DECLSPEC bool opal_ifisloopback(int if_index);
 
+/*
+ * Determine if a specified interface is included in a NULL-terminated argv array
+ */
+OPAL_DECLSPEC bool opal_ifmatches(int idx, char **nets);
 
 END_C_DECLS
 

orte/mca/oob/tcp/help-oob-tcp.txt

@@ -25,4 +25,11 @@ Dynamic ports:  %s
 
 Only one can be specified. Please choose either static or
 dynamic ports and try again.
+#
+[include-exclude]
+Both TCP interface include and exclude lists were specified:
+
+  Include: %s
+  Exclude: %s
 
+Only one of these can be given.

orte/mca/oob/tcp/oob_tcp.c

@@ -1299,39 +1299,65 @@ mca_oob_t* mca_oob_tcp_component_init(int* priority)
     int i;
     bool found_local = false;
     bool found_nonlocal = false;
+    char **interfaces = NULL;
+    mca_oob_tcp_device_t *dev;
+    bool including = true;
+    char name[32];
 
     *priority = 1;
 
     /* are there any interfaces? */
-    if(opal_ifcount() <= 0)
+    if (opal_ifcount() <= 0) {
         return NULL;
+    }
+
+    /* did someone mistakenly specify both includes AND excludes? */
+    if (NULL != mca_oob_tcp_component.tcp_include &&
+        NULL != mca_oob_tcp_component.tcp_exclude) {
+        orte_show_help("help-oob-tcp.txt", "include-exclude", true,
+                       mca_oob_tcp_component.tcp_include,
+                       mca_oob_tcp_component.tcp_exclude);
+        return NULL;
+    }
 
-    /* Which interfaces should we use?  Start by building a list of
-       all devices that meet the requirements of the if_include and
-       if_exclude list.  This might include local and non-local
-       interfaces mixed together. After that sorting is done, if there
-       is a mix of devices, we go through the devices that survived
-       the initial sort and remove all the local devices (since we
-       have non-local devices to use). */
+    /* if interface include was given, construct a list
+     * of those interfaces which match the specifications - remember,
+     * the includes could be given as named interfaces, IP addrs, or
+     * subnet+mask
+     */
+    if (NULL != mca_oob_tcp_component.tcp_include) {
+        interfaces = opal_argv_split(mca_oob_tcp_component.tcp_include, ',');
+        including = true;
+    } else if (NULL != mca_oob_tcp_component.tcp_exclude) {
+        interfaces = opal_argv_split(mca_oob_tcp_component.tcp_exclude, ',');
+        including = false;
+    }
+
+    /* look at all available interfaces */ 
     for (i = opal_ifbegin() ; i > 0 ; i = opal_ifnext(i)) {
-        char name[32];
-        mca_oob_tcp_device_t *dev;
 
+        /* get the name for diagnostic purposes */
         opal_ifindextoname(i, name, sizeof(name));
 
-        if (mca_oob_tcp_component.tcp_include != NULL &&
-            strstr(mca_oob_tcp_component.tcp_include,name) == NULL) {
-            OPAL_OUTPUT_VERBOSE((1, mca_oob_tcp_output_handle,
-                                 "%s oob:tcp:init rejecting interface %s",
-                                 ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), name));
-            continue;
-        }
-        if (mca_oob_tcp_component.tcp_exclude != NULL &&
-            strstr(mca_oob_tcp_component.tcp_exclude,name) != NULL) {
-            OPAL_OUTPUT_VERBOSE((1, mca_oob_tcp_output_handle,
-                                 "%s oob:tcp:init rejecting interface %s",
-                                 ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), name));
-            continue;
+        /* handle include/exclude directives */
+        if (NULL != interfaces) {
+            /* if we are including, then ignore this if not present */
+            if (including) {
+                if (!opal_ifmatches(i, interfaces)) {
+                    OPAL_OUTPUT_VERBOSE((1, mca_oob_tcp_output_handle,
+                                         "%s oob:tcp:init rejecting interface %s",
+                                         ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), name));
+                    continue;
+                }
+            } else {
+                /* we are excluding, so ignore if present */
+                if (opal_ifmatches(i, interfaces)) {
+                    OPAL_OUTPUT_VERBOSE((1, mca_oob_tcp_output_handle,
+                                         "%s oob:tcp:init rejecting interface %s",
+                                         ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), name));
+                    continue;
+                }
+            }
         }
 
         dev = OBJ_NEW(mca_oob_tcp_device_t);
@@ -1353,6 +1379,13 @@ mca_oob_t* mca_oob_tcp_component_init(int* priority)
         opal_list_append(&mca_oob_tcp_component.tcp_available_devices,
                          &dev->super);
     }
+
+    /* cleanup */
+    if (NULL != interfaces) {
+        opal_argv_free(interfaces);
+    }
+
+    /* remove all the local devices if we have non-local devices to use. */
     if (found_local && found_nonlocal) {
         opal_list_item_t *item, *next;
         for (item = opal_list_get_first(&mca_oob_tcp_component.tcp_available_devices) ;
@@ -1377,18 +1410,18 @@ mca_oob_t* mca_oob_tcp_component_init(int* priority)
     opal_hash_table_init(&mca_oob_tcp_component.tcp_peer_names, 128);
 
     opal_free_list_init(&mca_oob_tcp_component.tcp_peer_free,
-        sizeof(mca_oob_tcp_peer_t),
-        OBJ_CLASS(mca_oob_tcp_peer_t),
-        8,  /* initial number */
-        mca_oob_tcp_component.tcp_peer_limit, /* maximum number */
-        8);  /* increment to grow by */
+                        sizeof(mca_oob_tcp_peer_t),
+                        OBJ_CLASS(mca_oob_tcp_peer_t),
+                        8,  /* initial number */
+                        mca_oob_tcp_component.tcp_peer_limit, /* maximum number */
+                        8);  /* increment to grow by */
 
     opal_free_list_init(&mca_oob_tcp_component.tcp_msgs,
-        sizeof(mca_oob_tcp_msg_t),
-        OBJ_CLASS(mca_oob_tcp_msg_t),
-        8,  /* initial number */
-       -1,  /* maximum number */
-        8);  /* increment to grow by */
+                        sizeof(mca_oob_tcp_msg_t),
+                        OBJ_CLASS(mca_oob_tcp_msg_t),
+                        8,  /* initial number */
+                        -1,  /* maximum number */
+                        8);  /* increment to grow by */
 
 
     /* intialize event library */