lkml.org 
[lkml]   [2011]   [Oct]   [12]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [Xen-devel] [PATCH 3/3] xen/blk[front|back]: Enhance discard support with secure erasing support.
On Wed, Oct 12, 2011 at 11:54:02AM +0100, Ian Campbell wrote:
> On Tue, 2011-10-11 at 21:57 +0100, Konrad Rzeszutek Wilk wrote:
> > > My later response to it should include it:
> > > http://lists.xensource.com/archives/html/xen-devel/2011-10/msg00652.html
> > >
> > > >
> > > > Further I wonder why you don't use the "reserved" field instead of
> > > > extending the structure at the end.
> > >
> > > <blinks> I completly missed it. That would definitly work as well.
> > >
> > > Let me redo it with that in mind.
> >
> > I've posted the Xen hypervisor ABI one that thread above. The implementation
> > of that looks as follow:
>
> Ian.
>
> >
> > commit ae33f998d66c5982af533bda25c2b6c4f863789f
> > Author: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
> > Date: Mon Oct 10 10:58:40 2011 -0400
> >
> > xen/blk[front|back]: Enhance discard support with secure erasing support.
> >
> > Part of the blkdev_issue_discard(xx) operation is that it can also
> > issue a secure discard operation that will permanantly remove the
> > sectors in question. We advertise that we can support that via the
> > 'discard-secure' attribute and on the request, if the 'secure' bit
> > is set, we will attempt to pass in REQ_DISCARD | REQ_SECURE.
> >
> > CC: Li Dongyang <lidongyang@novell.com>
> > [v1: Used 'flag' instead of 'secure:1' bit]
> > [v2: Use 'reserved 'uint8_t' as a flag]
> > Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
> >
> > diff --git a/drivers/block/xen-blkback/blkback.c b/drivers/block/xen-blkback/blkback.c
> > index 94e659d..4f33c13 100644
> > --- a/drivers/block/xen-blkback/blkback.c
> > +++ b/drivers/block/xen-blkback/blkback.c
> > @@ -362,7 +362,7 @@ static int xen_blkbk_map(struct blkif_request *req,
> > {
> > struct gnttab_map_grant_ref map[BLKIF_MAX_SEGMENTS_PER_REQUEST];
> > int i;
> > - int nseg = req->nr_segments;
> > + int nseg = req->u1.nr_segments;
> > int ret = 0;
> >
> > /*
> > @@ -422,13 +422,16 @@ static void xen_blk_discard(struct xen_blkif *blkif, struct blkif_request *req)
> > int status = BLKIF_RSP_OKAY;
> > struct block_device *bdev = blkif->vbd.bdev;
> >
> > - if (blkif->blk_backend_type == BLKIF_BACKEND_PHY)
> > + if (blkif->blk_backend_type == BLKIF_BACKEND_PHY) {
> > + unsigned long secure = (blkif->vbd.discard_secure &&
> > + (req->u1.flag & BLKIF_OP_DISCARD_FLAG_SECURE)) ?
> > + BLKDEV_DISCARD_SECURE : 0;
> > /* just forward the discard request */
> > err = blkdev_issue_discard(bdev,
> > req->u.discard.sector_number,
> > req->u.discard.nr_sectors,
> > - GFP_KERNEL, 0);
> > - else if (blkif->blk_backend_type == BLKIF_BACKEND_FILE) {
> > + GFP_KERNEL, secure);
> > + } else if (blkif->blk_backend_type == BLKIF_BACKEND_FILE) {
> > /* punch a hole in the backing file */
> > struct loop_device *lo = bdev->bd_disk->private_data;
> > struct file *file = lo->lo_backing_file;
> > @@ -618,6 +621,9 @@ static int dispatch_rw_block_io(struct xen_blkif *blkif,
> > struct blk_plug plug;
> > bool drain = false;
> >
> > + /* Check that the number of segments is sane. */
> > + nseg = req->u1.nr_segments;
>
> This field is invalid (at least with these semantics) if req->operation
> == BLKIF_OP_DISCARD so reading it here and clearing it later when you
> decide it is invalid is just confusing. Why not read it inside the
> switch iff it is valid?

The problem was that 'nseg' would be read after the switch, so it would
contain the flag value. Which would throw off a lot of the loops which
would try to enumerate "(for (i = 0; i < nseg;...)".


Hence moving it to the top would make it valid for all the operations
except the BLKIF_OP_DISCARD. And BLKIF_OP_DISCARD would sensibly set it
nseg to zero so that we would not trip on those 'for (i = 0').

But I think you idea of making it an if statement would do, like:


> > @@ -643,8 +650,6 @@ static int dispatch_rw_block_io(struct xen_blkif *blkif,
> > break;
> > }
> >
if (operation != REQ_DISCARD)
/* Check that the number of segments is sane. */
nseg = req->nr_segments;
else
nseg = 0;

> > if (unlikely(nseg == 0 && operation != WRITE_FLUSH &&
> > operation != REQ_DISCARD) ||

And I guess we can also skip the REQ_DISCARD test here.

> > unlikely(nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST)) {

.. snip..
> handle isn't really only r/w, is it? If it is then I think we should
> just repeat the id fields within the union and pad so the offset is
> correct:
>
> struct blkif_request {
> uint8_t operation; /* BLKIF_OP_??? */
> union {
> struct {
> uint8_t nr_segments; /* number of segments */
> blkif_vdev_t handle;
> uint64_t id; /* private guest value, echoed in resp */
> blkif_sector_t sector_number;/* start sector idx on disk (r/w only) */
> struct blkif_request_segment seg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
> } rw;
> struct {
> uint8_t flags;
> blkif_vdev_t __pad; /* was "handle: only for read/write requests */
> uint64_t id; /* private guest value, echoed in resp */
blkif_sector_t sectore_number;
uint64_t nr_sectors;
> } discard;

I like that. So much easier to comprehend. Let me spin up a patch for that.


\
 
 \ /
  Last update: 2011-10-12 17:49    [W:0.096 / U:2.504 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site