> On Oct 5, 2024, at 12:51 PM, Pali Rohár <pali@xxxxxxxxxx> wrote: > > On Thursday 12 September 2024 19:34:02 Chuck Lever wrote: >> On Fri, Sep 13, 2024 at 01:28:20AM +0200, Pali Rohár wrote: >>> On Friday 13 September 2024 01:22:07 Pali Rohár wrote: >>>> On Friday 13 September 2024 09:10:45 NeilBrown wrote: >>>>> On Fri, 13 Sep 2024, Pali Rohár wrote: >>>>>> NLMv2 is completely different protocol than NLMv1 and NLMv3, and in >>>>>> original Sun implementation is used for RPC loopback callbacks from statd >>>>>> to lockd services. Linux does not use nor does not implement NLMv2. >>>>>> >>>>>> Hence, NLMv3 is not backward compatible with NLMv2. But NLMv3 is backward >>>>>> compatible with NLMv1. Fix comment. >>>>>> >>>>>> Signed-off-by: Pali Rohár <pali@xxxxxxxxxx> >>>>>> --- >>>>>> fs/lockd/clntxdr.c | 4 +++- >>>>>> 1 file changed, 3 insertions(+), 1 deletion(-) >>>>>> >>>>>> diff --git a/fs/lockd/clntxdr.c b/fs/lockd/clntxdr.c >>>>>> index a3e97278b997..81ffa521f945 100644 >>>>>> --- a/fs/lockd/clntxdr.c >>>>>> +++ b/fs/lockd/clntxdr.c >>>>>> @@ -3,7 +3,9 @@ >>>>>> * linux/fs/lockd/clntxdr.c >>>>>> * >>>>>> * XDR functions to encode/decode NLM version 3 RPC arguments and results. >>>>>> - * NLM version 3 is backwards compatible with NLM versions 1 and 2. >>>>>> + * NLM version 3 is backwards compatible with NLM version 1. >>>>>> + * NLM version 2 is different protocol used only for RPC loopback callbacks >>>>>> + * from statd to lockd and is not implemented on Linux. >>>>>> * >>>>>> * NLM client-side only. >>>>>> * >>>>> >>>>> Reviewed-by: NeilBrown <neilb@xxxxxxx> >>>>> >>>>> Do you have a reference for that info about v2? I hadn't heard of it >>>>> before. >>>>> >>>>> NeilBrown >>>> >>>> I have just this information in my notes. I guess it should be possible >>>> to gather more information about v2 from released Sun/Solaris source >>>> code via OpenSolaris / Illumos projects. >>> >>> Just very quickly I found this Illumos XDR file for NLM: >>> https://github.com/illumos/illumos-gate/blob/master/usr/src/uts/common/rpcsvc/nlm_prot.x >>> >>> And it defines NLMv2 with two procedures numbered 17 and 18, plus there >>> is a comment in file header about v2. >>> >>> So probably the best reference would be the Illumos source code. >> >> What you see in the Illumos code is not something that is part >> of the standard NLM protocol, but rather a private upcall protocol >> between the kernel and user space that is special sauce added >> by each implementation of NLM/NSM. > > Ok. But this applies for v2, no? On Linux, those operations are part of the NLMv1/3/4 protocol implementation, so essentially the NLM v2 functionality is a part of all NLM versions on Linux. >> Also note the way NLMv3 is defined in this file: it defines only >> a handful of new operations. The other operations are inherited >> from NLMv1. > > Yes, v3 is there and is inherited from v1. This is also what I pointed > in the comment. That v3 inherits from v1, not v2. Generally this is an abuse of the purpose of the RPC program versioning mechanism. Linux has a very similar upcall mechanism, but uses NLM procedure numbers that are set aside for this purpose instead of abusing a moribund protocol version. > In header file of that nlm_prot.x is written: > > * There are currently 3 versions of the protocol in use. Versions 1 > * and 3 are used with NFS version 2. Version 4 is used with NFS > * version 3. > * > * (Note: there is also a version 2, but it defines an orthogonal set of > * procedures that the status monitor uses to notify the lock manager of > * changes in monitored systems.) > > Which sounds like version 3 has nothing with version 2. > > My understanding of that comment is that version 2 contains only those > private upcall protocol between kernel and userspace about which you > wrote, and therefore version 3 is not backward compatible with version 2. > >> IMO the comment is accurate and does not warrant a change. How about this replacement: * XDR functions to encode/decode NLM version 1 and 3 RPC * arguments and results. NLM version 2 is not specified * by a standard, thus it is not implemented. -- Chuck Lever
