While I was digging into a dkms bug, I've noticed that kmod does not
fallback to the normal init_module() code path when the in-kernel
decompression fails.
As I looked closer, I realised that the current module compression
detection relies on kmod being build with zstd, xz, zlib ... Even in the
case where we rely on the kernel module itself to decompress - aka we
don't use the external libraries all together.
By building kmod without zstd and friends, we shave some space off the
initrd which helps with boot time.
Majority of this series handles the latter and fixes a few related bugs
along the way. The last commits is somewhat questionable commit and
addresses the former issue.
Happy to hear your feedback,
Emil
---
Emil Velikov (13):
libkmod: use a dup()'d fd for zlib
libkmod: keep gzFile gzf local to load_zlib()
libkmod: remove kmod_file::{zstd,xz}_used flags
libkmod: clear file->memory if map fails
libkmod: nuke struct file_ops
libkmod: propagate {zstd,xz,zlib}_load errors
libkmod: move kmod_file_load_contents as applicable
libkmod: always detect the module compression
libkmod: swap alloca usage for a few assert_cc
libkmod: tidy-up kmod_file_open()
libkmod: move load_reg() further up
libkmod: keep KMOD_FILE_COMPRESSION_NONE/load_reg in comp_types
libkmod: always fallback to do_init_module()
libkmod/libkmod-file.c | 206 ++++++++++++++++++---------------------------
libkmod/libkmod-internal.h | 2 +-
libkmod/libkmod-module.c | 8 +-
3 files changed, 90 insertions(+), 126 deletions(-)
---
base-commit: b29704cd448aaa455dba4e656fc0f0d3c686df3f
change-id: 20240212-decompression-fixes-56371eeb707d
Best regards,
--
Emil Velikov <emil.l.velikov@xxxxxxxxx>
