2.4.9-pre1 NFS problem

2.4.9-pre1 NFS problem

[Max Kamenetsky]

It looks like 2.4.9-pre1 breaks the NFS server.  Directories still get
exported and clients can mount them fine, but the ls command on the
client fails to report any files.  Filename completion also doesn't
work.  However, files can still be viewed if you know the exact file
name.

The same problem happens with 2.4.9-pre2.

Max

Re: 2.4.9-pre1 NFS problem

[Linus Torvalds]

In article <20010813114636.A4641@chinook.stanford.edu> you write:
>
>It looks like 2.4.9-pre1 breaks the NFS server.  Directories still get
>exported and clients can mount them fine, but the ls command on the
>client fails to report any files.  Filename completion also doesn't
>work.  However, files can still be viewed if you know the exact file
>name.

Yes, there's a missing off_t -> loff_t change in pre1, and the compile
will even warn about it...

>The same problem happens with 2.4.9-pre2.

..but I thought I fixed it in pre2. 

HOWEVER - there are a few others that I missed because the NFSD layer is
doing ugly casts of function pointers (don't ask me why - it should have
the right type in 'filldir_t' already but wants to use its own type), so
the compiler can't warn about it. 

Let's hear it for type safety and avoiding ugly casts.

Anyway, here's the patch - does this fix it for you?

		Linus

-----
diff -u --recursive --new-file pre2/linux/fs/nfsd/nfs3xdr.c linux/fs/nfsd/nfs3xdr.c
--- pre2/linux/fs/nfsd/nfs3xdr.c	Sun Aug 12 23:12:33 2001
+++ linux/fs/nfsd/nfs3xdr.c	Mon Aug 13 12:22:57 2001
@@ -730,14 +730,14 @@
 
 int
 nfs3svc_encode_entry(struct readdir_cd *cd, const char *name,
-		     int namlen, off_t offset, ino_t ino, unsigned int d_type)
+		     int namlen, loff_t offset, ino_t ino, unsigned int d_type)
 {
 	return encode_entry(cd, name, namlen, offset, ino, d_type, 0);
 }
 
 int
 nfs3svc_encode_entry_plus(struct readdir_cd *cd, const char *name,
-			  int namlen, off_t offset, ino_t ino, unsigned int d_type)
+			  int namlen, loff_t offset, ino_t ino, unsigned int d_type)
 {
 	return encode_entry(cd, name, namlen, offset, ino, d_type, 1);
 }
diff -u --recursive --new-file pre2/linux/fs/nfsd/nfsxdr.c linux/fs/nfsd/nfsxdr.c
--- pre2/linux/fs/nfsd/nfsxdr.c	Sun Aug 12 23:12:33 2001
+++ linux/fs/nfsd/nfsxdr.c	Mon Aug 13 12:23:49 2001
@@ -391,7 +391,7 @@
 
 int
 nfssvc_encode_entry(struct readdir_cd *cd, const char *name,
-		    int namlen, off_t offset, ino_t ino, unsigned int d_type)
+		    int namlen, loff_t offset, ino_t ino, unsigned int d_type)
 {
 	u32	*p = cd->buffer;
 	int	buflen, slen;
diff -u --recursive --new-file pre2/linux/include/linux/nfsd/nfsd.h linux/include/linux/nfsd/nfsd.h
--- pre2/linux/include/linux/nfsd/nfsd.h	Fri Aug 10 18:14:31 2001
+++ linux/include/linux/nfsd/nfsd.h	Mon Aug 13 12:30:35 2001
@@ -57,7 +57,7 @@
 	char			dotonly;
 };
 typedef int		(*encode_dent_fn)(struct readdir_cd *, const char *,
-						int, off_t, ino_t, unsigned int);
+						int, loff_t, ino_t, unsigned int);
 typedef int (*nfsd_dirop_t)(struct inode *, struct dentry *, int, int);
 
 /*
diff -u --recursive --new-file pre2/linux/include/linux/nfsd/xdr.h linux/include/linux/nfsd/xdr.h
--- pre2/linux/include/linux/nfsd/xdr.h	Fri Aug 10 18:15:05 2001
+++ linux/include/linux/nfsd/xdr.h	Mon Aug 13 12:28:54 2001
@@ -151,7 +151,7 @@
 int nfssvc_encode_readdirres(struct svc_rqst *, u32 *, struct nfsd_readdirres *);
 
 int nfssvc_encode_entry(struct readdir_cd *, const char *name,
-				int namlen, off_t offset, ino_t ino, unsigned int);
+				int namlen, loff_t offset, ino_t ino, unsigned int);
 
 int nfssvc_release_fhandle(struct svc_rqst *, u32 *, struct nfsd_fhandle *);
 
diff -u --recursive --new-file pre2/linux/include/linux/nfsd/xdr3.h linux/include/linux/nfsd/xdr3.h
--- pre2/linux/include/linux/nfsd/xdr3.h	Fri Aug 10 18:15:17 2001
+++ linux/include/linux/nfsd/xdr3.h	Mon Aug 13 12:28:54 2001
@@ -292,10 +292,10 @@
 int nfs3svc_release_fhandle2(struct svc_rqst *, u32 *,
 				struct nfsd3_fhandle_pair *);
 int nfs3svc_encode_entry(struct readdir_cd *, const char *name,
-				int namlen, off_t offset, ino_t ino,
+				int namlen, loff_t offset, ino_t ino,
 				unsigned int);
 int nfs3svc_encode_entry_plus(struct readdir_cd *, const char *name,
-				int namlen, off_t offset, ino_t ino,
+				int namlen, loff_t offset, ino_t ino,
 				unsigned int);
 
 

Re: 2.4.9-pre1 NFS problem

[Max Kamenetsky]

* Linus Torvalds <torvalds@transmeta.com> [08/13/01 18:40] wrote:
> In article <20010813114636.A4641@chinook.stanford.edu> you write:
> >It looks like 2.4.9-pre1 breaks the NFS server.  Directories still get
> >exported and clients can mount them fine, but the ls command on the
> >client fails to report any files.  Filename completion also doesn't
> >work.  However, files can still be viewed if you know the exact file
> >name.
> 
> Yes, there's a missing off_t -> loff_t change in pre1, and the compile
> will even warn about it...
> 
> >The same problem happens with 2.4.9-pre2.
> 
> ..but I thought I fixed it in pre2. 
> 
> HOWEVER - there are a few others that I missed because the NFSD layer is
> doing ugly casts of function pointers (don't ask me why - it should have
> the right type in 'filldir_t' already but wants to use its own type), so
> the compiler can't warn about it. 
> 
> Let's hear it for type safety and avoiding ugly casts.
> 
> Anyway, here's the patch - does this fix it for you?

[snip patch]

I didn't try the patch, but 2.4.9-pre3 fixes the problem.  I'm
assuming the patch is in that kernel version.

Max