[patch] Re: kern/94278: Crash with FIFOs and ktrace

[David Xu]

On Tuesday 14 March 2006 21:02, Kostik Belousov wrote:
> Sorry for garbled patch. I do not know why mutt decided to encode
> some "=" as =3D.
> 
> 
> Index: compat/linux/linux_misc.c
> ===================================================================
> RCS file: /usr/local/arch/ncvs/src/sys/compat/linux/linux_misc.c,v
> retrieving revision 1.172
> diff -u -r1.172 linux_misc.c
> --- compat/linux/linux_misc.c	28 Dec 2005 07:08:54 -0000	1.172
> +++ compat/linux/linux_misc.c	14 Mar 2006 11:45:57 -0000
> @@ -310,6 +310,21 @@
>  	 * XXX: This should use vn_open() so that it is properly authorized,
>  	 * and to reduce code redundancy all over the place here.
>  	 */
> +	if (vp->v_type == VLNK) {
> +		error = EMLINK;
> +		goto cleanup;
> +	}
> +	if (vp->v_type == VSOCK) {
> +		error = EOPNOTSUPP;
> +		goto cleanup;
> +	}
> +	if (vp->v_type == VFIFO) {
> +		/* Due to way fifo works (by overloading f_ops),
> +		 * tricking kernel into write to the fifo leads to
> +		 * panic. Make a band-aid to filter the case. */
> +		error = EOPNOTSUPP;
> +		goto cleanup;
> +	}
>  #ifdef MAC
>  	error = mac_check_vnode_open(td->td_ucred, vp, FREAD);
>  	if (error)
> Index: fs/fifofs/fifo_vnops.c
> ===================================================================
> RCS file: /usr/local/arch/ncvs/src/sys/fs/fifofs/fifo_vnops.c,v
> retrieving revision 1.132
> diff -u -r1.132 fifo_vnops.c
> --- fs/fifofs/fifo_vnops.c	1 Oct 2005 20:15:41 -0000	1.132
> +++ fs/fifofs/fifo_vnops.c	14 Mar 2006 11:46:07 -0000
> @@ -168,6 +168,7 @@
>  		int  a_mode;
>  		struct ucred *a_cred;
>  		struct thread *a_td;
> +		int a_fdidx;
>  	} */ *ap;
>  {
>  	struct vnode *vp = ap->a_vp;
> Index: kern/vfs_syscalls.c
> ===================================================================
> RCS file: /usr/local/arch/ncvs/src/sys/kern/vfs_syscalls.c,v
> retrieving revision 1.411
> diff -u -r1.411 vfs_syscalls.c
> --- kern/vfs_syscalls.c	4 Mar 2006 00:09:09 -0000	1.411
> +++ kern/vfs_syscalls.c	14 Mar 2006 11:46:10 -0000
> @@ -4101,6 +4101,13 @@
>  		error = EOPNOTSUPP;
>  		goto bad;
>  	}
> +	if (vp->v_type == VFIFO) {
> +		/* Due to way fifo works (by overloading f_ops),
> +		 * tricking kernel into write to the fifo leads to
> +		 * panic. Make a band-aid to filter the case. */
> +		error = EOPNOTSUPP;
> +		goto bad;
> +	}
>  	mode = 0;
>  	if (fmode & (FWRITE | O_TRUNC)) {
>  		if (vp->v_type == VDIR) {
> Index: kern/vfs_vnops.c
> ===================================================================
> RCS file: /usr/local/arch/ncvs/src/sys/kern/vfs_vnops.c,v
> retrieving revision 1.238
> diff -u -r1.238 vfs_vnops.c
> --- kern/vfs_vnops.c	11 Mar 2006 17:14:05 -0000	1.238
> +++ kern/vfs_vnops.c	14 Mar 2006 11:46:10 -0000
> @@ -194,6 +194,13 @@
>  		error = EOPNOTSUPP;
>  		goto bad;
>  	}
> +	if ((vp->v_type == VFIFO) && (fdidx < 0)) {
> +		/* Due to way fifo works (by overloading f_ops),
> +		 * tricking kernel into write to the fifo leads to
> +		 * panic. Make a band-aid to filter the case. */
> +		error = EOPNOTSUPP;
> +		goto bad;
> +	}
>  	mode = 0;
>  	if (fmode & (FWRITE | O_TRUNC)) {
>  		if (vp->v_type == VDIR) {
> 

I know, someone will work out such a messy patch, but is it reasonable ?
why does not the fifi code suddenly work with well defined vnode interface ?
why did someone want to break the well defined FILE->vnode->fs->device
layers ? sigh.

David Xu