Coverity CID 187785

Use slurm_seterrno(), rather than maintaining a plugin-local and
credential local version of that errno.

Remove g_slurm_auth_errno() and g_slurm_auth_errstr(). Replace all
calling locations with %m format identifiers which should pick up
the errno in that thread, which is now directly set by the auth plugin.

The one place that this represents a slight regression in functionality
is that we loose the specific EMUNGE errno values. However, the places
that will set this in the auth/munge plugin are already printing it
through an error() with munge_ctx_strerror(). Replace those two locations
with ESLURM_AUTH_CRED_INVALID instead.

While currently safe, this is blocking further refactoring, and
much less conceptually clean than testing the cred properly with
g_slurm_auth_verify().

Stop fetching a lot of (now) unused fields, and print the two
relevant lines. Drop errors to warnings - we know the decode
failed, so failure to retrieve the encode and decode time is
quite likely.

Unused, and until earlier in this commit series the auth/munge
plugin was printing to stdout directly rather than the fp argument.

Move the message for ESLURM_AUTH_UNPACK into slurm_auth.c, which
removes all error message handling except the complicated MUNGE handing
from the plugins. Adjust MUNGE_ERRNO_OFFSET far enough to avoid the
range these have just been renumbered into.

Use a new cred_wrapper_t struct designed to overlap the opaque cred
structures used within the individual plugins. Add the index value
to the start of each of these structures.

The index  is set when a given credential is unpacked, and used to lookup
the appropriate plugin for use with the remaining auth plugin
API calls.

The zeroth element is always the AuthType, and will be used by default.

This is currently only supported in the slurmctld and slurmdbd;
ignore this in all other calling locations.

Hard code to using the first plugin loaded for now.

Expose the plugin_type symbol from the plugins.

Ignore the plugin_version symbol, since on the unpack side we never do
anything with it. Just send a zero instead, and continue to ignore it
on the unpack side.

Note that for auth/munge, the 19.05 and the older protocol blocks are
identical, so collapse them back into one block.

The plugin_id will be sent in first by g_auth_pack, so this
check here is now redundant and can be removed to simplify things.

Add a protocol_version dependent block to g_auth_pack / g_auth_unpack.

For the 19.05 protocol_version, send the plugin_id first, then call
unpack routines inside the individual plugins. Throw an error if there
is a mismatch for now.

Update slurm.spec and slurm.spec-legacy as well.

Update slurm.spec and slurm.spec-legacy as well.

Revert "Avoid errors from packing accounting_storage_mysql.so when RPM is built with out mysql support."

This reverts commit d18e26b9.

Backport of 21b9a39b

Bug 6625

Update slurm.spec and slurm.spec-legacy as well.

Update slurm.spec and slurm.spec-legacy as well.

No changes in logic

The plugin-specific errno handling these were intended for does
not exist, and these ranges can be reused without issue at this point.

No plugins implement this interface, so discard.

Never called. None of the plugins have bothered to implement this correctly.
The jobcomp/script plugin comes closest, but did not pick error codes within
the designated range.

Remove "cons_tres: " from some src/common/gres.c log messages.
They are vestigial from relocated functions.

This needs to be replaced by a finer-grained access control in the
next pre-release of v19.05