On Fri, Jul 25, 2014 at 09:43:37AM -0400, Jeffrey Altman wrote: > On 7/25/2014 4:38 AM, Nathaniel W Filardo wrote: > > Salutations -devel@, > > > > I have three related sets of questions for you. > > > > First, conflicting serverFlags bits: > > > > Which of afs/vldbint.h or afs/volser.h do I believe? They agree on "new > > repsite", "rw", "ro", and "backup" volumes, but then one says "rwrepl", > > "dontuse" while the other says "uuid", "dontuse", "rwrepl": > > > > afs/volser.h > >> #define NEW_REPSITE 0x01 > >> #define ITSROVOL 0x02 > >> #define ITSRWVOL 0x04 > >> #define ITSBACKVOL 0x08 > >> #define ITSRWREPL 0x10 > >> #define RO_DONTUSE 0x20 > > These are used within the volserver and in the RPCs to the volserver. > > > afs/vldbint.h > >> #define VLSF_NEWREPSITE 0x01 /* flags for indiv. server entry */ > >> #define VLSF_ROVOL 0x02 > >> #define VLSF_RWVOL 0x04 > >> #define VLSF_BACKVOL 0x08 > >> #define VLSF_UUID 0x10 > >> #define VLSF_DONTUSE 0x20 > >> #define VLSF_RWREPLICA 0x40 /* Volume is a RW replica */ > > These are used within the vlserver and in the RPCs to the vlserver.
But src/volser/vsprocs.c:/^UV_ReleaseVolume, for example, says:
> vcode = VLDB_GetEntryByID(afromvol, RWVOL, &entry);
>[...]
> for (i = 0; i < entry.nServers; i++) {
> if (entry.serverFlags[i] & ITSROVOL) {
>[...]
and src/volser/vsutils.c:/^VLDB_GetEntryByID calls ubik_VL_GetEntryByID,
which looks to be a vlserver RPC? Forgive me if I'm still missing
something.
> > Second, VLOP flags:
> >
> > While here, why are VLOP flags defined in vldbint.h with self-styled bogus
> > clashing names in volser.h?
>
> The original set of VLOP flags do not cover all desired situations but
> they cannot be extended because otherwise an unknown value could be sent
> to a client that doesn't recognize it. As a result overloading of
> values was performed. New VLOP_ macros were created for the new
> operations for code clarity.
I was more asking about the location of the definitions than the clashing.
> > And third, please send help, I am so confused:
> >
> > I believe the following assertions, could someone please tell me their
> > validity?
> >
> > 1) A clone volume is identified by the cloneId field, which may only be
> > set if the RO_EXISTS bit is on. There's not really a separate entry
> > for the clone volume; the semantics are inherited from the one with
> > VLSF_ROVOL turned on.
>
> There is a special readonly temporary clone that is used as part of vos
> release. Other clones can be created but they do not get recorded in
> the VLDB.
OK. Could you please look at the logic in
http://gerrit.openafs.org/#patch,sidebyside,10966,7,src/volser/vos-foreach.c
line 245 and tell me that it's not crazy?
> > 2) VLSF_BACKVOL is not used in practice; the VLDB marks backup volumes
> > with VLSF_RWVOL.
>
> Backup volumes must exist on the same site (server/partition) as the RW
> volume. At one point in history long ago this might not have been true.
That does not really answer the question. What is VLSF_BACKVOL indicating,
if anything? It is asserted only in src/libafscp/afscp_volume.c and seems
to get some play in src/vlserver/vlprocs.c:/^SVL_ListAttributesN2 (but to
what end I cannot follow) and is consulted in
src/vlserver/vldb_check.c:/^readentry . ITSBACKVOL has a similarly sparse
occurrence throughout the code. src/volser/vsprocs.c:/^CheckVolume will
assert it (and pass the resulting entry to a VLDB RPC!);
src/bucoord/commands.c:/^EvalVolumeSet2 checks for it, and
src/libadmin/vos/afs_vosAdmin.c:/^copyVLDBEntry checks for it.
Thanks again,
--nwf;
pgpC0gbQNwB64.pgp
Description: PGP signature
