On Tue, Jan 30, 2007 at 03:21:19PM -0800, Andrew Morton wrote: > On Tue, 30 Jan 2007 12:55:58 -0800 > Andrew Morton <[EMAIL PROTECTED]> wrote: > > > y'know, four or five years back I fixed this bug by doing > > > > current->locked_page = page; > > > > in the write() code, and then teaching the pagefault code to avoid locking > > the same page. Patch below. > > > > But then evil mean Hugh pointed out that the patch is still vulnerable to > > ab/ba deadlocking so I dropped it. > > And he was right, of course. Task A holds file a's i_mutex and takes a > fault against file b's page. Task B holds file b's i_mutex and takes a > fault against file a's page. Drat. > > I wonder if there's a sane way of preventing that.
If you want to go down the path of carrying state around in task_struct, you can take the mmap_sem and set a flag, then get_user_pages the source page and lock both source and destination in ascending order, then your page fault handler checks the flag and skips mmap_sem, and the rest of your fault path checks both the page locks you're holding. At which point you arrive at a horrible mess :) - To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
