diff options
Diffstat (limited to 'README')
| -rw-r--r-- | README | 87 |
1 files changed, 87 insertions, 0 deletions
@@ -0,0 +1,87 @@ + + About + ***** + +This directory contains the source code of a collection of tools for working +with SquashFS file systems. + +The `gensquashfs` program takes as input a file listing similar to the +program `gen_init_cpio` in the Linux kernel source tree and produces a +SquashFS image. + +The input list precisely specifies the directory structure, all permission bits +and all UIDs & GIDs of who owns what. + +The tool doesn't care if the directories, symlinks, device special files, +etc... actually exist when packing a SquashFS image. The only thing that +really has to exist are the input _files_ that can be placed arbitrarily in +the file system by specifying input and target locations. + +All directory entries are sorted by name and processed sequentially. All time +stamps in the SquashFS image are set to a command line specified value (or 0 +by default). Thus the entire process should be deterministic, i.e. same input +produces byte-for-byte the same output. + + +The `rdsquashfs` program can read a SquashFS image and produce file listings, +extract individual files or entire sub trees of the image to a desired +location. + + + Why not use the official squashfs-tools? + **************************************** + +The mksquashfs utility is semi-broken and generally a PITA to work with. + +For the typically use case of SquashFS (i.e. as rootfs for a live distro or an +embedded system), it should be blindingly obvious that I might want to micro +manage what goes into the file system, that UIDs/GIDs of the host system are +garbage inside the image and that setting the desired permissions (e.g. suid) +or SELinux labels on the input is completely out of the question. Also, it +would be really cool if the whole thing was reproducible. + +All of this seems to have been an afterthought with mksquashfs. Some of it can +be achieved with exclusion options, "pseudo files" or "filters" (a completely +undocumented feature that not even `--help` tells you about). + +My main gripes with mksquashfs were the following: + + - I need to precisely replicate the entire filesystem for packing, even tough + the only thing actually needed are in theory the regular files. + - Files in the input FS but not in the pseudo file are still packed + but with garbage UID/GID from the host system. + - When I want files that are not owned by root, the root inode will get + garbage UID/GID from the host system and there is no way to change this. + - mtime is read from the input file system and there is no way to override it. + - Data is packed by a thread pool, i.e. in a non-deterministic way. + - Extended attributes are read from the input file system, i.e. the only way + to get SELinux labels into the SquashFS filesystem is to set them on the + input data. + +That's at least what I can think of right now from the top of my head. + +It would be preferable to fix mksquashfs itself, but the source code is a +horrid dumpster fire. It turned out to be easier to understand the structure +of SquashFS by reading the available documentation plus kernel source and +implementing the desired feature set from scratch. + +Furthermore, upstream seems to be unmaintained at the moment and the mailing +list appears to be about as dead as SourceForge that hosts it. + + + Limitations + *********** + +The entire code base is at the moment fairly fresh and has been hacked together +in a weekend or two. So naturally, the feature set it implements is currently +quite limited. + +At the moment, the following things are still missing: + + - extended attributes + - sparse files + - hard links + - NFS export tables + - compressor options + - compressors other than XZ and GZIP (i.e. lzo, lz4, zstd, *maybe* lzma1?) + - support for extracting SquashFS < 4.0 |