On Mon, 18 Aug 2025 20:40:23 +0200, Nicolas Schier wrote: > On Thu, Aug 14, 2025 at 03:17:58PM +1000, David Disseldorp wrote: > > This patchset adds copy_file_range() support to gen_init_cpio. When > > combined with data segment alignment, large-file archiving performance > > is improved on Btrfs and XFS due to reflinks (see 7/7 benchmarks). > > > > cpio data segment alignment is provided by "bending" the newc spec > > to zero-pad the filename field. GNU cpio and Linux initramfs extractors > > handle this fine as long as PATH_MAX isn't exceeded. > > > > Changes since v1 RFC > > - add alignment patches 6-7 > > - slightly rework commit and error messages > > - rename l->len to avoid 1/i confusion > > > > David Disseldorp (7): > > gen_init_cpio: write to fd instead of stdout stream > > gen_init_cpio: support -o <output_path> parameter > > gen_init_cpio: attempt copy_file_range for file data > > gen_init_cpio: avoid duplicate strlen calls > > gen_initramfs.sh: use gen_init_cpio -o parameter > > docs: initramfs: file data alignment via name padding > > gen_init_cpio: add -a <data_align> as reflink optimization > > > > .../driver-api/early-userspace/buffer-format.rst | 5 + > > usr/gen_init_cpio.c | 234 ++++++++++++++------- > > usr/gen_initramfs.sh | 7 +- > > 3 files changed, 166 insertions(+), 80 deletions(-) > > > > > > Thanks for the series! I have found only a minor nick pick and some few > bike-shedding things. > > Reviewed-by: Nicolas Schier <nsc@xxxxxxxxxx> Thanks a lot for the review, Nicolas! I'll post a v3 with your suggestions addressed, as I have one further initramfs_test.c patch to padded filename test coverage. Cheers, David
