gtfs: Remove the EOF optimization

Turns out this is wrong for directories (and often device files).  The
ufs server reports a directory size of 4096, which is what Linux tells
it.  We'd read 4096 bytes worth of records, and then stop.

The 9p man page says:

	 The length records the number of bytes in the file. Directories
	 and most files representing devices have a conventional length
	 of 0.

Instead of mucking around, we'll just issue the RPC for whatever offset
we're asked for.

Signed-off-by: Barret Rhoden <brho@cs.berkeley.edu>

kern/drivers/dev/gtfs.c

diff --git a/kern/drivers/dev/gtfs.c b/kern/drivers/dev/gtfs.c
index a7284ba..bd375a0 100644
--- a/kern/drivers/dev/gtfs.c
+++ b/kern/drivers/dev/gtfs.c
@@ -46,7 +46,10 @@
  * be_{length,mode,mtime} should be what the remote server thinks they are -
  * especially for length and mode.  The invariant is that e.g. the file's length
  * == be_length, and the qlock protects that invariant.  We don't care as much
- * about mtime, since some 9p servers just change that on their own. */
+ * about mtime, since some 9p servers just change that on their own.
+ *
+ * Also note that you can't trust be_length for directories.  You'll often get
+ * 4096 or 0, depending on the 9p server you're talking to. */
 struct gtfs_priv {
 	struct chan					*be_walk;	/* never opened */
 	struct chan					*be_read;
@@ -323,10 +326,6 @@
 {
 	struct gtfs_priv *gp = fsf_to_gtfs_priv(f);
 
-	if (off >= gp->be_length) {
-		/* We can skip the RPC, since we know it will return zero (EOF). */
-		return 0;
-	}
 	if (!gp->be_read)
 		gp->be_read = cclone_and_open(gp->be_walk, O_READ);
 	return devtab[gp->be_read->type].read(gp->be_read, ubuf, n, off);