On Tue, Mar 22, 2011 at 10:17 AM, Jeff Layton <jlayton@xxxxxxxxxx> wrote:
> On Tue, 22 Mar 2011 09:50:43 -0500
> Steve French <smfrench@xxxxxxxxx> wrote:
>
>> On Tue, Mar 22, 2011 at 7:04 AM, Jeff Layton <jlayton@xxxxxxxxxx> wrote:
>> > On Tue, 22 Mar 2011 15:01:11 +0300
>> > Pavel Shilovsky <piastry@xxxxxxxxxxx> wrote:
>> >
>> >> 2011/3/22 Jeff Layton <jlayton@xxxxxxxxxx>:
>> >> > On Wed, 16 Mar 2011 01:55:32 +0300
>> >> > Pavel Shilovsky <piastry@xxxxxxxxxxx> wrote:
>> >> >
>> >> >> Use invalidate_inode_pages2 that don't leave pages even if shrink_page_list()
>> >> >> has a temp ref on them. It prevents a data coherency problem when
>> >> >> cifs_invalidate_mapping didn't invalidate pages but the client thinks that a data
>> >> >> from the cache is uptodate according to an oplock level (exclusive or II).
>> >> >>
>> >> >> Signed-off-by: Pavel Shilovsky <piastry@xxxxxxxxxxx>
>> >> >> ---
>> >> >> fs/cifs/inode.c | 10 ++++++++--
>> >> >> 1 files changed, 8 insertions(+), 2 deletions(-)
>> >> >>
>> >> >> diff --git a/fs/cifs/inode.c b/fs/cifs/inode.c
>> >> >> index 41e5651..adb6324 100644
>> >> >> --- a/fs/cifs/inode.c
>> >> >> +++ b/fs/cifs/inode.c
>> >> >> @@ -1691,12 +1691,18 @@ cifs_invalidate_mapping(struct inode *inode)
>> >> >>
>> >> >> cifs_i->invalid_mapping = false;
>> >> >>
>> >> >> - /* write back any cached data */
>> >> >> if (inode->i_mapping && inode->i_mapping->nrpages != 0) {
>> >> >> + /* write back any cached data */
>> >> >> rc = filemap_write_and_wait(inode->i_mapping);
>> >> >> mapping_set_error(inode->i_mapping, rc);
>> >> >> + rc = invalidate_inode_pages2(inode->i_mapping);
>> >> >> + if (rc) {
>> >> >> + cERROR(1, "%s: could not invalidate inode %p", __func__,
>> >> >> + inode);
>> >> >> + cifs_i->invalid_mapping = true;
>> >> >> + }
>> >> >> }
>> >> >> - invalidate_remote_inode(inode);
>> >> >> +
>> >> >> cifs_fscache_reset_inode_cookie(inode);
>> >> >> }
>> >> >>
>> >> >
>> >> > Hi Pavel,
>> >> >
>> >> > I noticed that Steve has merged this patch, but it doesn't seem like
>> >> > you ever made the change to have EBUSY percolate up to userspace. Are
>> >> > you still planning to fix that?
>> >> >
>> >>
>> >> Steve NACK'ed this and we came to agreement to mark a mapping as
>> >> invalid in this case for now and let the client try invalidate it
>> >> again when it needs. Of course, it can bring problems with a data
>> >> coherency if you need to use a cache very strictly.
>> >>
>> >
>> > Ahh, I must have missed the discussion around that NAK, and I don't see
>> > any record of it in the list archives. Steve, can you clarify why you
>> > didn't think that approach was acceptible?
>>
>> Pavel and I discussed the code paths in detail via IRC or chat. I
>> would have to go back and look through the code again, but I thought
>> the issue was that returning
>> EBUSY (from invalidate_inode_pages2) doesn't make any sense to
>> revalidate_file (e.g. why would a seek return EBUSY even though we
>> successfully updated the correct file size?) who can't use the rc, but
>> by resetting the mapping to invalid_mapping invalidate_inode_pages2
>> will still get invoked again later (which is all we would do in the
>> caller if we passed it back).
>>
>> In practice it probably doesn't make much difference for cifs anyway
>> (passing the return code back from invalidate_inode_pages2) because we
>> don't have a "launder_page" operation for cifs and therefore don't
>> return an error, and for the other case it doesn't seem correct to
>> fail a seek because of (page_has_private(page))
>>
>
> It does however make sense to return an error if you're going to
> invalidate a page and can't because the page is dirty:
>
> ret2 = do_launder_page(mapping, page);
> if (ret2 == 0) {
> if (!invalidate_complete_page2(mapping, page))
> ret2 = -EBUSY;
> }
>
> do_launder_page returns 0 if there is no launder_page op...
>
> invalidate_complete_page does this:
>
> if (PageDirty(page))
> goto failed;
>
> ...so the fact that we don't have a launder_page operation doesn't mean
> that we'll never return an error here.
>
> It makes no sense at all to do a cERROR printk for this. It'll make
> users go "huh?", and won't help us to debug anything.
but how could we have PageDirty? We just did filemap_write_and_wait
(unless some private mmap case where it may be ok?).
--
Thanks,
Steve
--
To unsubscribe from this list: send the line "unsubscribe linux-cifs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
