So far ramfs uses equally sized chunks. This series changes it to use dynamically sized chunks instead. When making a file bigger ramfs now always tries to put the additional data in a single chunk. With this we get fewer chunks and with it better performance as we have less list iterations to do. Also this allows us to provide a memmap implementation for ramfs which works for a good bunch of cases. When a file is created, truncated to its final size and then filled with data, the data will end up in one contiguous memory region. In this case we can provide a pointer to this data which is important for usecases where we would otherwise duplicate big image files in memory. Sascha Sascha Hauer (11): update list.h from Linux-5.7 fs: Add destroy_inode callbacks to filesystems fs: Make iput() accept NULL pointers fs: free inodes we no longer need digest: Drop usage of memmap fs: ramfs: Return -ENOSPC fs: ramfs: Drop dead code fs: ramfs: Use dynamically sized chunks fs: ramfs: Implement memmap libfile: copy_file: Fix calling discard_range libfile: copy_file: explicitly truncate to final size crypto/digest.c | 27 +- fs/cramfs/cramfs.c | 10 + fs/devfs.c | 8 + fs/fs.c | 7 +- fs/nfs.c | 8 + fs/ramfs.c | 349 ++++++++++--------- fs/squashfs/squashfs.c | 10 +- include/linux/compiler.h | 1 - include/linux/list.h | 700 ++++++++++++++++++++++++++++++--------- include/linux/poison.h | 26 ++ include/linux/types.h | 12 + lib/libfile.c | 25 +- 12 files changed, 826 insertions(+), 357 deletions(-) create mode 100644 include/linux/poison.h -- 2.27.0 _______________________________________________ barebox mailing list barebox@xxxxxxxxxxxxxxxxxxx http://lists.infradead.org/mailman/listinfo/barebox
