lfs changelog --user cl5 testfs-MDT0000 fails for a changelog user that was
registered plainly. The same command works if the user was registered with a
mask or a name.

Registration and lookup disagree about the record type.

mdd_changelog_user_reg() deliberately writes the old record type when the
caller gave neither a mask nor a name, for compatibility
(lustre/mdd/mdd_device.c:1785-1791):

        /* keep old record type for users without mask/name for
         * compatibility needs
         */
        if (mask || (name && name[0]))
                rec->cur_hdr.lrh_type = CHANGELOG_USER_REC2;
        else
                rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;

mdd_changelog_user_lookup_cb() accepts only the new one
(lustre/mdd/mdd_changelog.c:46):

        if ((rec->cur_hdr.lrh_type != CHANGELOG_USER_REC2) ||
            (req->cf_user_id != 0 && rec->cur_id != req->cf_user_id) ||
            (req->cf_user_id == 0 && strcmp(rec->cur_name, req->cf_username)))
                RETURN(0);

So the record is skipped, mcul_found stays 0, and
mdd_changelog_user_lookup() turns that into -ENOENT
(lustre/mdd/mdd_changelog.c:100-103).
Userspace reports cannot set changelog filter.

It is the only walker in MDD that does this. Every other one takes both types
and guards the fields that exist in only one of them:

        mdd_device.c:248-249   changelog_user_init_cb()           accepts both, warns otherwise
        mdd_device.c:333-334   changelog_user_detect_orphan_cb()  accepts both, warns otherwise
        mdd_device.c:222       mdd_chlg_username()                reads cur_name only for REC2
        mdd_device.c:233       mdd_chlg_usermask()                returns 0 for the old type

The path is the shipped command, not only the library call:

        lfs changelog --user cl5 testfs-MDT0000
          -> llapi_changelog_start_user()       liblustreapi_chlg.c:204
          -> ioctl(OBD_IOC_CHANGELOG_FILTER)    liblustreapi_chlg.c:230
          -> mdd_changelog_user_lookup()        mdd_changelog.c:76
          -> mdd_changelog_user_lookup_cb()     mdd_changelog.c:29   -> ENOENT

Reproducer

        lctl --device testfs-MDT0000 changelog_register        # -> cl5
        lctl --device testfs-MDT0000 changelog_register -m all # -> cl6
        lfs changelog --user cl6 testfs-MDT0000                # works
        lfs changelog --user cl5 testfs-MDT0000                # ENOENT

Note that deregistering the last changelog user purges the log, so a script
that cleans up after itself reads an empty changelog on the next run. That is
expected behaviour, not part of this bug.

The fix

Accept CHANGELOG_USER_REC as well, and guard the two REC2-only fields
rather than the whole record.

This is not a one-word change, because struct llog_changelog_user_rec is
half the size of struct llog_changelog_user_rec2 and has neither
cur_mask nor cur_name (lustre_idl.h:3084-3104). Today the ||
short-circuits on the type test, so rec->cur_name is never reached for an
old record; simply dropping that first clause would read past the end of the
record into whatever the llog buffer holds next. So:

• a lookup by ID must match an old record on cur_id alone;
• a lookup by name cannot match an old record at all, since it has no name, and should skip it rather than compare;
• the reply's mask comes from mdd_chlg_usermask(), which already returns 0 for the old type, and the reply's username must stay empty for it.

How it was found

On a lab run of the LU-20462 changelog scanner on 2026-08-25, where a plainly
registered cl5 failed and cl6, registered with -m, worked. That
scanner reaches the changelog through the same ioctl, so it hit the same wall.

Introduced by 5b85a4eb75 (LU-19296, "changelog: retrive changelog user info
from MDT", October 2025), which added the lookup; the registration side was
already writing both types.