[Dnsmasq-discuss] Enable dnsmasq to return addresses of multiple tftp servers in cycle

[Simon Kelley]

Sushil Agrawal <sushil_agrawal at hotmail.com> wrote:

>
>
>Hi All,
>
>The attached patch enables dnsmasq(8) to return the ip addresses
>of multiple tftp servers (for pxe boot) having a single alias
>name (thru the /etc/hosts) in round robin fashion.
>
>We have a virtual environment where there are some managed hosts
>running kvm hypervisor and some > 150  virtual machines
>carved out of these hosts. All these virtual machines are thin
>clients which boot of the net using pxe. The arrangement is such
>that all of these machines boot at the same time and in order to
>balance the load on the tftp server, we planned to have multiple
>tftp servers (on these managed hosts themselves) running.
>
>dnsmasq currently returns only one address for tftp server
>which it finds in the dhcp-boot option in the dnsmasq.conf
>file. Although there is a provision for specifying the
>tftp server name there but it is not used. We made
>use of this name as an alias for multiple tftp servers
>(using /etc/hosts) and then return the resolved addresses
>one by one in cycle.
>
>The function cache_find_by_name() already takes care
>of all this and all we had to do was to just call
>it in dhcp_reply() to get one address and return
>that as tftp address.
>
>If the tftp server address is specified in the dhcp-boot
>option then that address is returned always. If address
>is not specified and name is specified then this cyclic
>stuff is done. This solution is working for us and hope it
>will be useful to others.
>
>
>Following is the patch (also attached) based on
>dnsmasq-2.57:
>
>................................................................
>
>--- dnsmasq-2.57.origg/src/rfc2131.c    2011-02-17 21:00:15.000000000
>+0530
>+++ dnsmasq-2.57/src/rfc2131.c    2011-06-18 17:48:59.000000000 +0530
>@@ -829,8 +829,23 @@
>              and set discovery_control = 8 */
>           if (boot)
>             {
>-              if (boot->next_server.s_addr)
>+              if (boot->next_server.s_addr) {
>             mess->siaddr = boot->next_server;
>+              } else if (boot->sname) {
>+
>+                /*
>+                 * If the pxe server name is given in the conf then
>+                 * cycle thru its addresses.
>+                 */
>+
>+                struct crec *crecp;
>+                time_t now = dnsmasq_time();
>+                crecp = cache_find_by_name(NULL, boot->sname, now,
>+                               F_DHCP | F_HOSTS);
>+                if (crecp) {
>+                    mess->siaddr = *(struct in_addr*)&crecp->addr;
>+                }
>+              }
>               
>               if (boot->file)
>         strncpy((char *)mess->file, boot->file, sizeof(mess->file)-1);
>@@ -2222,8 +2237,23 @@
>         strncpy((char *)mess->file, boot->file, sizeof(mess->file)-1);
>     }
>       
>-      if (boot->next_server.s_addr)
>+      if (boot->next_server.s_addr) {
>     mess->siaddr = boot->next_server;
>+      } else if (boot->sname) {
>+
>+        /*
>+         * If the pxe server name is given in the conf then
>+         * cycle thru its addresses.
>+         */
>+
>+        struct crec *crecp;
>+          time_t now = dnsmasq_time();
>+        crecp = cache_find_by_name(NULL, boot->sname, now,
>+                       F_DHCP | F_HOSTS);
>+        if (crecp) {
>+            mess->siaddr = *(struct in_addr*)&crecp->addr;
>+        }
>+      }
>     }
>   else
>    /* Use the values of the relevant options if no dhcp-boot given and
>................................................................
>
>Sushil.
>
>
> 		 	   		  _______________________________________________
>Dnsmasq-discuss mailing list
>Dnsmasq-discuss at lists.thekelleys.org.uk
>http://lists.thekelleys.org.uk/mailman/listinfo/dnsmasq-discuss

This looks like a great idea, but overloading the server name parameter is not correct. That can be used for other things. The correct way to do this is to check if a name is supplied instead of an address in dhcp-boot and act accordingly. That will need some code in options.c and elaboration of the dhcp boot data structure. 

A general observation for patch submitters. I'm much more likely to look kindly on a patch if it includes documentation, ie a patch to the man page.

I'm away from home at the moment, so further action on this may have to wait until next week.

Cheers, 

Simon.