Linux kernel mirror (for testing) git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel os linux

docs: initramfs: file data alignment via name padding

The existing cpio extraction logic reads (maximum PATH_MAX) name_len
bytes from the archive into the collected name buffer and ensures that
the trailing byte is a null-terminator. This allows the actual file name
to be shorter than name_len, with the name string simply zero-terminated
prior to the last byte.

Initramfs generators, such as dracut-cpio[1], can take advantage of name
zero-padding to align file data segments within the archive to
filesystem block boundaries. Block boundary alignment may allow the
copy_file_range syscall to reflink archive source and destination
extents.

Link: https://github.com/dracutdevs/dracut/commit/300e4b116c624bca1b9e7251708b1ae656fe9157 [1]
Signed-off-by: David Disseldorp <ddiss@suse.de>
Reviewed-by: Nicolas Schier <nsc@kernel.org>
Link: https://lore.kernel.org/r/20250819032607.28727-7-ddiss@suse.de
Signed-off-by: Nathan Chancellor <nathan@kernel.org>

authored by

David Disseldorp and committed by
Nathan Chancellor
7c1f14f6 9135564d

+5
+5
Documentation/driver-api/early-userspace/buffer-format.rst
··· 86 86 The c_filesize should be zero for any file which is not a regular file 87 87 or symlink. 88 88 89 + c_namesize may account for more than one trailing '\0', as long as the 90 + value doesn't exceed PATH_MAX. This can be useful for ensuring that a 91 + subsequent file data segment is aligned, e.g. to a filesystem block 92 + boundary. 93 + 89 94 The c_chksum field contains a simple 32-bit unsigned sum of all the 90 95 bytes in the data field. cpio(1) refers to this as "crc", which is 91 96 clearly incorrect (a cyclic redundancy check is a different and