Page MenuHome GnuPG

Mismatch between enums and config table
Open, HighPublic


In gpgconf.h, the enum gc_component_id_t has a comment:

/* Component system.  Each component is a set of options that can be
 * configured at the same time.  If you change this, don't forget to
 * update gc_component[] in gpgconf-comp.c.  */

it defines an enum value GC_COMPONENT_TPM2DAEMON (6) and GC_COMPONENT_PINENTRY (8).

In gpgconf-comp.c, the table gc_component looks like this:

} gc_component[GC_COMPONENT_NR] =
   /* Note: The order of the items must match the order given in the
    * gc_component_id_t enumeration.  The order is often used by
    * frontends to display the backend options thus do not change the
    * order without considering the user experience.  */
   { NULL },   /* DUMMY for GC_COMPONENT_ANY */

but the component-data for the TPM2 daemon is included only if BUILD_WITH_TPM2D is defined. So when that define is off, the gc_component table is one shorter than intended. This has two effects:

  • read after the end of the table when looping in gc_component_retrieve_options() .
  • mismatch between IDs and the actual entries; in particular, entry 6 in the table is the one for dirmngr, and entry 7 is now pinentry. Code that loops and checks to skip pinentry (enum value 8) now hits pinentry accidentally, as entry 7 in the table.

This yields downstream bug report -- gpg tries to collect options all components, skipping pinentry; the code skips enum value 8, but reaches pinentry in table entry 7 already.

I applied a simple patch that adds

     { NULL }, /* Another dummy, to keep the enum in sync with this table */

so there is a null entry in the table at place 6 (where the tpm daemon goes) and the table's length and entries are in-sync with the enum again.


git master

Event Timeline

So that you don't need to chase the downstream bug report, the problem from a user's perspective looks like this:

$ gpgconf --apply-defaults
gpgconf: WARNING: invalid line in option table of '/usr/local/bin/pinentry'

gpgconf: name of config file for /usr/local/bin/pinentry is not known

gpgconf: fatal error (exit status 1)

The Warning line comes from retrieve_options_from_program() in ./tools/gpgconf-comp.c, because pinentry doesn't understand --dump-option-table as a command-line-argument and responds with OK nice to meet you. The relevant loop that goes over the table of components is in gc_component_retrieve_options() in the same file.

werner triaged this task as High priority.Tue, Nov 23, 1:29 PM
werner added a subscriber: werner.

Thanks for the well written bug report and the fix.