Re: nfs mount fail

[Ingo Molnar <mingo@elte.hu>]

* Trond Myklebust <trond.myklebust@fys.uio.no> wrote:

> > --- a/fs/nfs/super.c
> > +++ b/fs/nfs/super.c
> > @@ -1253,6 +1253,7 @@ static int nfs_parse_mount_options(char *raw,
> >  			default:
> >  				dfprintk(MOUNT, "NFS:   unrecognized "
> >  						"transport protocol\n");
> > +				kfree(string);
> >  				return 0;
> >  			}
> >  			break;
> 
> There is a possible clean up there too. We can move the other kfree() 
> calls out of the inner switch statement, and coalesce them all into a 
> single call.

Correct - separately from the leak fix. (which potentially wants to go 
to -stable as well)

Plus it's not just the kfree() calls that can be refactored but also the 
~25 match_strdup() call sites. Most of these repetitive sequences:

                case Opt_retrans:
                        string = match_strdup(args);
                        if (string == NULL)
                                goto out_nomem;
                        rc = strict_strtoul(string, 10, &option);
                        kfree(string);
                        if (rc != 0 || option == 0)
                                goto out_invalid_value;
                        mnt->retrans = option;
                        break;

could be pushed into a helper function, along the lines of:

                case Opt_retrans:
			if (parse_opt(args, &mnt->retrans) < 0 || mnt->retrans == 0)
                                goto out_invalid_value;
                        break;

where the non-repetitive value checks can be done after a generic 
parse_opt(). (or something like that)

That makes it more readable as well, as the switch statement will only 
list true per option properties, with minimal repetitive patterns.

	Ingo