Messages in this thread | | | Date | Mon, 27 Jan 2014 20:49:00 -0700 | From | Matthew Wilcox <> | Subject | Re: [PATCH v5 22/22] XIP: Add support for unwritten extents |
| |
On Mon, Jan 27, 2014 at 04:32:07PM -0700, Ross Zwisler wrote: > It looks like we have an additional bit of complexity with the hole case. The > issue is that for holes, bh->b_size is just the full size of the write as set > earlier in the function: > > bh->b_size = ALIGN(end - offset, PAGE_SIZE); > > >From this code it seems like you hoped the call into get_block() would adjust > bh->b_size to the size of the hole, allowing you to zero just the hole space > in the user buffer. It doesn't look like it does, though, at least for ext4.
Argh. I got confused. ext4 *has* this information, it just doesn't propagate it into the bh if it's a hole! Should it? The comments in the direct IO code imply that it *may*, but doesn't *have* to. What it's doing now (not touching it) is probably wrong.
> To just assume the current FS block is a hole, we can do something like this:
Yes, this should fix things on an interim basis. Bit inefficient, but it'll work.
> diff --git a/fs/xip.c b/fs/xip.c > index 35e401e..e902593 100644 > --- a/fs/xip.c > +++ b/fs/xip.c > @@ -122,7 +122,7 @@ static ssize_t xip_io(int rw, struct inode *inode, const struct > > if (hole) { > addr = NULL; > - size = bh->b_size - first; > + size = (1 << inode->i_blkbits) - first; > } else { > retval = xip_get_addr(inode, bh, &addr); > if (retval < 0) > > -- > To unsubscribe from this list: send the line "unsubscribe linux-ext4" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html
-- Matthew Wilcox Intel Open Source Technology Centre "Bill, look, we understand that you're interested in selling us this operating system, but compare it to ours. We can't possibly take such a retrograde step."
| |