OK... I see now... ce23e64 (Al's patch) got merged to mainline after
2dcd0af (Linux 4.6) and Andreas' Orangefs patches won't load
without ce23e64. I was working from vanilla Linux 4.6. My little
O_DIRECT patch is also affected, since .direct_IO lost the offset
argument.
So... now the top of my local tree is ea8ea73, Andreas' patch
series loads fine, I've fixed my O_DIRECT patch, it loads, and
the whole thing compiles and installs great. But won't boot. On my VM.
Even when built without Andreas' patch series and my O_DIRECT patch
it won't boot. So now I've got it booted with the vanilla 4.6 kernel.
I guess this is a "tree of the moment" problem. I guess Linus' tree
will continue to evolve and I'll come in tomorrow
and fetch from it again and it will boot. And then I can load
and test the patches, update the Orangefs linux-next and
build the pull request before the merge window ends.
Maybe after six or eight more merge windows I'll get the
hang of being upstream and quit causing trouble <g>...
-Mike
On Thu, May 26, 2016 at 4:05 PM, Al Viro <viro@xxxxxxxxxxxxxxxxxx> wrote:
> On Thu, May 26, 2016 at 03:44:23PM -0400, Mike Marshall wrote:
>> Hi Andreas...
>>
>> I've looked over your patch series, and was about
>> to apply it and run some tests... but the first patch
>> "orangefs: Remove useless defines" won't apply.
>>
>> I received your patch series a day or so after I merged
>> our kernel.org tree up to 4.6, and at first I thought maybe
>> you'd composed your patch during the long spell when
>> the last commit in our tree was e56f49814250f4ca4b66ec7d3a71152846761d1b,
>> resulting in some kind of conflict, but that's not it. Now
>> that we're upstream, you're probably patching against
>> the "real" Linux tree anyhow...
>>
>> Here's where the fail happens:
>>
>> @@ -470,7 +470,7 @@ static int orangefs_xattr_get_default(const struct
>> xattr_handler *handler,
>> size_t size)
>> {
>> return orangefs_inode_getxattr(inode,
>> - ORANGEFS_XATTR_NAME_DEFAULT_PREFIX,
>> + "",
>> name,
>> buffer,
>> size);
>>
>> "return orangefs_inode_getxattr(inode," should be
>> "return orangefs_inode_getxattr(dentry->d_inode,"
>>
>> I looked way back in time, and don't think it ever
>> was "inode" instead of "dentry->d_inode", or how
>> it could ever be.
>
> Mainline has ->getxattr() called with dentry and inode passed separately;
> moreover, dentry is *NOT* guaranteed to be already associated with the
> inode. See commit ce23e64 and, subsequently, b96809.
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
