On Sat, Mar 07, 2009 at 08:59:16PM +0200, Benny Halevy wrote: > On Mar. 06, 2009, 23:32 +0200, "J. Bruce Fields" <bfields@xxxxxxxxxxxx> wrote: > > On Thu, Mar 05, 2009 at 12:09:40PM +0200, Benny Halevy wrote: > >> On Mar. 05, 2009, 11:32 +0200, Ni Wenjuan <niwj@xxxxxxxxxxxxxx> wrote: > >>> the result of newpynfs test case of LINK4a . if you link with target directoty > >>> is a symbole file,it should get NFS4ERR_NOTDIR ,instead got NFS4ERR_SYMLINK. > >>> > >>> THE LINK operation don't list NFS4ERR_SYMLINK as valid errors in the spec. But > >>> NFS4ERR_SYMLINK seems like a reasonable error. Is this an oversight > >>> in the spec, or something we need to fix? > >> Although NFSv4.1 adds NFS4ERR_SYMLINK to LINK's allowed errors list > >> (and this might be an indication for it being an oversight in rfc3530), > > > > The error lists in rfc3530 are known to be incomplete in some cases, so > > before adding an exception like this I'd like something more. (E.g.: > > does this cause any client or application to fail? Is there some > > logical reason notdir is a more useful error than symlink?) > > FWIW, the linux nfs client translates NFS4ERR_SYMLINK to -ELOOP > which is awkward and less descriptive to the app / user than > -ENOTDIR. Hm, OK. If we fix this will -ELOOP then become reasonable for our remaining NFS4ERR_SYMLINK returns? > That said, I don't think a careful client implementation > should ever get NFS4ERR_SYMLINK if it stats the directory it operates > on before sending the link op (or lookup, create, rename, etc.) to make > sure it is indeed a directory, right?. That sounds racy. --b. -- To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html