Re: jffs_file_write always appends to file.

[Finn Håkansson <finn@xxxxxxx.com>]

Hi David,

This patch uses the ideas in your patch but takes care of
one more bug as you can see. I also would like to remove
the call to jffs_delete_data() that you added because that
is supposed to be done from jffs_insert_node() later on.

With this patch I got the correct result when I ran your
test program.

Thanks a lot!

/Finn


Index: inode-v22.c
===================================================================
RCS file: /home/cvs/mtd/fs/jffs/inode-v22.c,v
retrieving revision 1.14
diff -u -r1.14 inode-v22.c
--- inode-v22.c 2000/08/02 13:34:46     1.14
+++ inode-v22.c 2000/08/02 19:50:40
@@ -1440,7 +1440,7 @@
                goto out;
        }
        DJM(no_jffs_node++);
-       node->data_offset = f->size;
+       node->data_offset = pos;
        node->removed_size = 0;
 
        /* Initialize the raw inode.  */
@@ -1459,7 +1459,7 @@
        raw_inode.atime = CURRENT_TIME;
        raw_inode.mtime = raw_inode.atime;
        raw_inode.ctime = f->ctime;
-       raw_inode.offset = f->size;
+       raw_inode.offset = pos;
        raw_inode.dsize = count;
        raw_inode.rsize = 0;
        raw_inode.nsize = 0;
@@ -1467,6 +1467,11 @@
        raw_inode.spare = 0;
        raw_inode.rename = 0;
        raw_inode.deleted = 0;
+
+       if (pos < f->size) {
+               raw_inode.rsize = jffs_min(count, f->size - pos);
+               node->removed_size = raw_inode.rsize;
+       }
 
        /* TODO: BAAAAAAAAD! buf is a userspace-pointer, and should be
                 treated as such, with copy_from_user etc...  */



On Wed, 2 Aug 2000, David Woodhouse wrote:

> This patch makes it use the 'ppos' argument, but even with the addition of 
> the jffs_delete_data() call, the filesystem needs to be remounted before 
> the 'new' data appear in the right position.
> 
>  (... more stuff that might be fixed if we were using the page cache, 
> methinks...)
> 
> I'm off home. Bob, if you want to take a look at this, feel free. This 
> patch is all there is that I haven't committed to CVS. I posted my test 
> program earlier, IIRC.
> 
> Index: inode-v22.c
> ===================================================================
> RCS file: /home/cvs/mtd/fs/jffs/inode-v22.c,v
> retrieving revision 1.14
> diff -u -w -r1.14 inode-v22.c
> --- inode-v22.c	2000/08/02 13:34:46	1.14
> +++ inode-v22.c	2000/08/02 16:32:46
> @@ -1459,7 +1459,7 @@
>  	raw_inode.atime = CURRENT_TIME;
>  	raw_inode.mtime = raw_inode.atime;
>  	raw_inode.ctime = f->ctime;
> -	raw_inode.offset = f->size;
> +	raw_inode.offset = pos;
>  	raw_inode.dsize = count;
>  	raw_inode.rsize = 0;
>  	raw_inode.nsize = 0;
> @@ -1467,6 +1467,11 @@
>  	raw_inode.spare = 0;
>  	raw_inode.rename = 0;
>  	raw_inode.deleted = 0;
> +
> +	if (pos < f->size) {
> +		node->removed_size = raw_inode.rsize = jffs_min(count, f->size - pos);
> +		jffs_delete_data(f, node);
> +	}
>  
>  	/* TODO: BAAAAAAAAD! buf is a userspace-pointer, and should be
>  	         treated as such, with copy_from_user etc...  */
> 
> 
> --
> dwmw2
> 
> 
>