On Thu, Jul 26, 2018 at 08:21:43PM +0800, Gao Xiang wrote: > Hi, > > This is actually the 2nd patchset of erofs file system, > the original patchset can be found at > > Link: https://marc.info/?l=linux-fsdevel&m=152776480425624 > > In order to keep up with the mainline linux-kernel changes and > improve it in a more active and timely manner, we put forword > this upstream proposal for linux-staging. > > EROFS file system is a read-only file system with compression > support designed for certain devices (especially embeded > devices) with very limited physical memory and lots of memory > consumers, such as Android devices. It aimes to provide > a complete compression solution for such devices focuing > on high performance and little extra memory overhead. > > It is perferred to select larger compressed cluster sizes > (generally >= 128k) for traditional compression file systems. > It reads and decompresses a large compressed cluster at once, > which has a good-looking random read number when memory > is sufficient because all historial decompressed data > is expected to be cached in memory. However, it also > induces destructive effects when such devices have no enough > spare memory for caching and decompression. > > EROFS file system acts in some different way. It uses > fixed-sized compressed size rather than fixed-sized input > size, namely VLE (variable-length extent) compression, > which has at least three adventages: > > 1) all data read from block device at once can be > utilized, and read amplification can be easier to > estimate and control; > 2) generally, it has a better compression ratio than > fixed-sized input compression approaches configured > with the same size; > 3) aggressively optimized paths such as partial page > read can be implemented to gain better performance > for page-unaligned read (unimplemented yet, in TODO list). > > As can be seen, VLE compression does a great job in small > compressed cluster sizes, which is of course suitable for > devices with limited memory. In this patchset, an in-place > decompresion is also introduced to minimize extra memory usage. > > > Apart from compression, EROFS also has the following features > available and some limits: > o page-sized block support (currently, and no buffer-head); > o 32-bit block address (16TB for 4KB block); > o selectable v1 (32 bytes) / v2 (64 bytes) inode; > o 32-bit / 64-bit file size; > o 64-bit node number for addressing inodes; > o 64-bit s and 32-bit ns timestamps; > o inline data support; > o inline and shared xattr support; > o metadata and data can be mixed (optional); > o special inode support; > o posix acl support. > > > The file system is still actively WIP, see _TODO_ for more details. > > Any comments are welcome. :) Thanks for submitting this, the filesystem looks very interesting. I've queued it all up now in the staging-next tree. greg k-h _______________________________________________ devel mailing list devel@xxxxxxxxxxxxxxxxxxxxxx http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel
