On Tue, Nov 05, 2019 at 09:12:33AM -0500, Eli Schwartz wrote:
On 11/5/19 9:03 AM, Allan McRae wrote:
On 5/11/19 11:58 pm, Morten Linderud wrote:
On Tue, Nov 05, 2019 at 11:54:34PM +1000, Allan McRae wrote:
When creating or modifying repo tarballs, place a .TIMESTAMP file with seconds since epoch in it. This will be used in the future to enable rejecting databases older that a given threshold.
Also skip reading the .TIMESTAMP file in sync_db_populate().
Signed-off-by: Allan McRae <allan@archlinux.org> ---
<snip>
diff --git a/scripts/repo-add.sh.in b/scripts/repo-add.sh.in index caf1232d..c87409f1 100644 --- a/scripts/repo-add.sh.in +++ b/scripts/repo-add.sh.in @@ -526,6 +526,7 @@ create_db() { TAR_OPT=$(verify_repo_extension "$REPO_DB_FILE") # $LOCKFILE is already guaranteed to be absolute so this is safe dirname=${LOCKFILE%/*} + timestamp=$(date +%s)
This should probably utilize SOURCE_DATE_EPOCH or something equivalent?
timestamp=$(date --date="@${SOURCE_DATE_EPOCH:-$(date +%s)}" +%s))
Why? I can see no reason why it should...
I don't either see value in "reproducible builds" for the actual state of the database. It's just a series of plaintext pointers to some other (hopefully reproducible) packages.
If we actually did want to respect SOURCE_DATE_EPOCH, we'd need to do a lot more, like doing that for the bsdtar metadata (both file timestamps and file owners, probably sort files too, etc.) but again, I don't see how this protects the supply chain.
Hmm, should probably discuss the threat model or attack vectors in #archlinux-reproducible. -- Morten Linderud PGP: 9C02FF419FECBE16