lkml.org 
[lkml]   [2007]   [Jan]   [11]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    Patch in this message
    /
    From
    SubjectRE: [patch] optimize o_direct on block device - v3
    Date
    Andrew Morton wrote on Thursday, January 11, 2007 11:29 AM
    > On Thu, 11 Jan 2007 13:21:57 -0600
    > Michael Reed <mdr@sgi.com> wrote:
    > > Testing on my ia64 system reveals that this patch introduces a
    > > data integrity error for direct i/o to a block device. Device
    > > errors which result in i/o failure do not propagate to the
    > > process issuing direct i/o to the device.
    > >
    > > This can be reproduced by doing writes to a fibre channel block
    > > device and then disabling the switch port connecting the host
    > > adapter to the switch.
    > >
    >
    > Does this fix it?
    >
    > <thwaps Ken>


    Darn, kicking myself in the butt. Thank you Andrew for fixing this.
    We've also running DIO stress test almost non-stop over the last 30
    days or so and we did uncover another bug in that patch.

    Andrew, would you please take the follow bug fix patch as well. It
    is critical because it also affects data integrity.


    [patch] fix blk_direct_IO bio preparation.

    For large size DIO that needs multiple bio, one full page worth of data
    was lost at the boundary of bio's maximum sector or segment limits.
    After a bio is full and got submitted. The outer while (nbytes) { ... }
    loop will allocate a new bio and just march on to index into next page.
    It just forget about the page that bio_add_page() rejected when previous
    bio is full. Fix it by put the rejected page back to pvec so we pick it
    up again for the next bio.


    Signed-off-by: Ken Chen <kenneth.w.chen@intel.com>

    diff -Nurp linux-2.6.20-rc4/fs/block_dev.c linux-2.6.20.ken/fs/block_dev.c
    --- linux-2.6.20-rc4/fs/block_dev.c 2007-01-06 21:45:51.000000000 -0800
    +++ linux-2.6.20.ken/fs/block_dev.c 2007-01-10 19:54:53.000000000 -0800
    @@ -190,6 +190,12 @@ static struct page *blk_get_page(unsigne
    return pvec->page[pvec->idx++];
    }

    +/* return a pge back to pvec array */
    +static void blk_unget_page(struct page *page, struct pvec *pvec)
    +{
    + pvec->page[--pvec->idx] = page;
    +}
    +
    static ssize_t
    blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
    loff_t pos, unsigned long nr_segs)
    @@ -278,6 +284,8 @@ same_bio:
    count = min(count, nbytes);
    goto same_bio;
    }
    + } else {
    + blk_unget_page(page, &pvec);
    }

    /* bio is ready, submit it */
    -
    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to majordomo@vger.kernel.org
    More majordomo info at http://vger.kernel.org/majordomo-info.html
    Please read the FAQ at http://www.tux.org/lkml/
    \
     
     \ /
      Last update: 2007-01-11 23:17    [W:2.171 / U:0.656 seconds]
    ©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site