mirror of
https://github.com/containers/youki
synced 2026-10-21 02:20:24 +02:00
Emilio PerssonandClaude Opus 5.5
78cdc04103
* fix(libcontainer): mount overlays whose lowerdir exceeds 255 bytes The fd-based mount flow passes each mount option to fsconfig(FSCONFIG_SET_STRING). The kernel copies the value with strndup_user(value, 256) and returns EINVAL for anything longer than 255 bytes (fs/fsopen.c). Podman's image volumes (`--mount type=image`) are read-only overlay mounts whose lowerdir names two layers under the storage root; for rootless Podman that value is typically close to 300 bytes, so the container fails with "failed to prepare rootfs". When an overlay lowerdir value is too long for one fsconfig string, pass the layers one at a time as "lowerdir+" (and "datadir+" after a "::" separator), which overlayfs accepts since Linux 6.8. Shorter values are passed unchanged. Signed-off-by: Emilio Persson <93738720+ItzExotical@users.noreply.github.com> Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(libcontainer): keep find_parent_mount's doc comment on it The new helper was added between find_parent_mount and its doc comment. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: Emilio Persson <93738720+ItzExotical@users.noreply.github.com> * fix(libcontainer): fall back to mount(2) when fsconfig() fails Instead of splitting a long overlay lowerdir into "lowerdir+" layers (Linux 6.8+), fall back to the legacy mount(2) for a mount whose fsconfig() fails: it takes a lowerdir over fsconfig's 255-byte limit and options that older kernels' fsconfig() rejects. A tmpcopyup mount keeps the error, since the copy-up needs the detached mount. Signed-off-by: Emilio Persson <93738720+ItzExotical@users.noreply.github.com> Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Signed-off-by: Emilio Persson <93738720+ItzExotical@users.noreply.github.com> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>