microfs - Minimally Improved Compressed Read Only File System

Origin: https://bitbucket.org/erikedlund/linux-microfs

Mirror: https://github.com/edlund/linux-microfs

microfs is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty
of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
See the GNU General Public License for more details.

microfs is basically a reimplementation of cramfs with a few new features added which are not backwards compatible. It is mostly implemented to serve as an exercise for its author and might not be particularly useful in any real world scenario. If you are looking for a cramfs replacement, you probably want to consider squashfs (which is an excellent read only file system). With that said, you might still find microfs interesting, particularly if you appreciate cramfs for its awesome simplicity.

The difference between cramfs and microfs lies in that microfs

Always store metadata as little endian and convert data to the endianess of the host when it is necessary/required.
Support configurable block sizes (ranging from 512 bytes up to 1 megabyte).
Support image sizes larger than 272 mb (the upper limit is 2^32 bytes).
Support files sizes larger than 16 mb (the upper limit is 2^32 - 1 bytes).
Support slightly longer file names (3 extra bytes).
Support ctime/mtime values for VFS inodes by using the ctime value for the image stored in the superblock.
Will try to uncompress data from the buffer heads directly to the page cache pages if it is possible.
Will have different read buffers for different mounted images in order to try to maximize the number of cache hits.
Support different options for dealing with concurrent requests for decompression of data.
Does not support file holes by simply skipping the zero bytes like cramfs, instead it will treat file holes as any other data and compress it.
Split the image into four main parts; 1) superblock, 2) inodes and dentries, 3) block pointers and 4) the compressed blocks. See the section "image format" for more information.
Support LZ4, LZO, XZ and ZSTD as alternatives to ZLIB.

Licensing

Most of microfs is licensed under the GPL. There are however a few utilities which might be useful outside of the project and are licensed under the new BSD license. Check the copyright and license notice at the top of the file to know what license a specific file is under. GPL will apply if no license is explicitly specified in a file.

Linux kernel compatibility

microfs is mainly focused on supporting a small number of kernel versions for popular distrubutions in the debian family. This is mostly because the upstream kernel is such a fast moving target that it would require extra effort to develop against.

Supported dists/kernels at the time of writing:

v0.2, Ubuntu: 4.10.0-x-generic (17.04, "Zesty Zapus")
v0.3, Ubuntu: 4.15.0-x-generic (18.04, "Bionic Beaver")

It may well work with other versions and other dists, but that has not been tested.

Image format

microfs images has four main parts:

super block/decompressor data
inodes/dentries
block pointers
compressed data blocks

The main difference compared to cramfs is that the block pointers are placed in their own section rather than being placed just before the compressed data blocks that they point to. The idea is that this change should allow microfs to use its buffers more efficiently. The downside is that each regular file and symlink requires one extra block pointer with this setup.

It is possible that decompressor specific data is stored immediately after the super block. This will happen if the decompressor needs some extra information that is not normally available, one example is the dictionary size for the xz decompressor.

Building microfs

It is presumed that you are planning to build microfs on a GNU/Linux system which can compile the Linux kernel. Adding to the requirements implied by this assumption microfs will also need the following things in order to compile successfully:

gcc (>=4.7), http://gcc.gnu.org/
GNU make (>=3.81), http://www.gnu.org/software/make/
GNU bash (>=4.2), http://www.gnu.org/software/bash/
check unit test framework (>=0.9.8), http://check.sourceforge.net/
zlib (>=1.2.7), http://www.zlib.net/
liblz4 (>=r94), http://code.google.com/p/lz4/ (opt)
liblzo (>=2.06), http://www.oberhumer.com/opensource/lzo/ (opt)
liblzma (>=5.2), https://www.7-zip.org/sdk.html (opt)
libzstd (>=1.3), https://github.com/facebook/zstd (opt)
python (>=2.7), http://www.python.org/
perl (>=5.14), http://www.perl.org/
cramfs-tools (>=1.1), http://sourceforge.net/projects/cramfs/ (opt)
squashfs-tools (>=4.2), http://squashfs.sourceforge.net/ (opt)
inotify-tools (>=3.14), https://github.com/rvoicilas/inotify-tools (opt)
tmpfs (>=3.11.0), https://www.kernel.org/ (opt)

(To get a quick and dirty dev environment up and running, you can try make devenv, which will try to figure out what Linux dist you are running and install the necessary packages.)

Once the build environment is set up it is sufficient to run make from the root source directory to build the lkm and the hostprogs with ZLIB support.

It is possible to control which compression libraries are supported by specifying LIB_*-params for make. Available options are:

LIB_LZ4
LIB_LZO
LIB_XZ
LIB_ZLIB
LIB_ZSTD

For example, to add support for LZ4, simply use make LIB_LZ4=1. Please note that the hostprogs will always need zlib in order to compile, so specifying LIB_ZLIB=0 will not allow microfs to compile in an environment where ZLIB is not installed.

Use the make command line argument DEBUG=1 or the combination DEBUG=1 DEBUG_SPAM=1 to do a debug build. Please note that a debug spam lkm will spam your syslog with messages (hence the name), it can however be very useful when dealing with a bug which is reproducible with a small set of files (or few operations).

Mount options

It is possible to tweak the behaviour of microfs by specifying options when an image is mounted.

metadata_blkptrbufsz=%u: The desired size of the metadata buffer for block pointers.
metadata_dentrybufsz=%u: The desired size of the metadata buffer for dentries/inodes.
decompressor_data_creator=%s: How microfs should handle decompressor data. See the section "Decompressor data" below. Valid values are: singleton, percpu, queue.
decompressor_data_acquirer=%s: How microfs should acquire an instance of decompressor data. See the section "Decompressor data" below. Valid values are: private, public.
debug_mountid=%u: Specify a mount ID which can help with debugging. It will printed as an INFO log message when the image is mounted.
debug_cksig: Check the super block signature, this is normally not required as the magic number is always checked.

An example: mount -o decompressor_data_creator=queue ....

Decompressor data

"decompressor data" is basically what a specific decompressor needs to be able to handle compressed blocks. It is usually internal state and buffers for the decompressor in question. This topic is a little complicated as microfs offers many options to allow its user to tweak how it is handled. It is documented in microfs.h, look for the doctag for struct microfs_decompressor_data to learn more. The default behaviour will (hopefully) be good enough for most people.

Testing microfs

Reproducible "randomness"

Pseudorandomness plays a very important role in the functional testing of microfs. Most of the data used for the tests is pseudorandomly generated and can be regenerated at any time if the seed that was used in the first place is known. This should provide a couple of major advantages:

microfs is tested on a wide variety of pseudorandom trees of files and directories.
Test sessions should be reproducible even if the actual data that is used is pseudorandom.
Random data will compress very badly, meaning that the test suite also can serve as a stress test of sorts.

Remote checks

microfs supports "remote checks", which makes it easy to run the test suite on some machine other than your primary workstation. Using a virtual machine is probably the best solution as it is very easy to reset and easier to debug if (when) something goes terribly wrong.

An example:

$ make remotecheck \
>     REMOTEHOST="localhost" \
>     REMOTEPORT="2222" \
>     REMOTEUSER="erik" \
>     REMOTEDEST="~" \
>     REMOTEMKARGS="CHECKARGS='-r 1387916272'"

Port 2222 on localhost is forwarded to a virtual machine with a ssh server.

Of the given arguments, REMOTEMKARGS is extra interesting. Use it to pass command line arguments to the remote make invocation. In the above example test.sh is seeded with the value 1387916272. Another example of using REMOTEMKARGS could be to set it to DEBUG=1 DEBUG_SPAM=1 in order to do a debug build.

Stress/Load testing

Use the flag "-S" for test.sh in order to have it run a stress and/or load test on the lkm. The idea is to try to test that the lkm works as it should even when the system is under heavy load and have little RAM to work with.

The contrast is an ordinary test run which pretty much allows the lkm to run under ideal circumstances. Hopefully there will not be too many bugs hiding between the extremes.

$ make remotecheck \
>     REMOTEHOST="localhost" \
>     REMOTEPORT="2222" \
>     REMOTEUSER="erik" \
>     REMOTEDEST="~" \
>     REMOTEMKARGS="CHECKARGS='-r 1389994471 -S'"

"Quick" tests

Use the flag "-Q" for test.sh in order to limit the block sizes tested to 512, 4096, 131072 and 1048576 (otherwise test.sh will use block sizes 512, 1024, 2048, 4096, 8192, 16384, 32768, 65536, 131072 and 1048576).

Raspberry PI (Raspbian)

The tests should pass without any flags, but there is a risk that sun might turn into a red giant before any test results are displayed. It is possible to wait a little less by running a quick test, reducing the size budget and changing the checksum algorithm:

$ make remotecheck \
>     REMOTEHOST="192.168.x.y" \
>     REMOTEPORT="22" \
>     REMOTEUSER="pi" \
>     REMOTEDEST="~" \
>     REMOTEMKARGS="CHECKARGS='-r 1389994471 -b 67108864 -C md5sum -Q'"

One should probably avoid stress testing.

Performance

Obviously every block size tested larger than PAGE_SIZE will mostly be interesting to compare microfs with squashfs, cramfs will still use PAGE_SIZE sized blocks.

The result presented by the benchmark tools is the average time that a test took over N passes. The idea is that the average should give a more fair representation of the performance of the filesystem than a single pass would.

Still, with all this said, it is still probably wise to consider the bundled benchmark tools to be biased towards making microfs look good, even if that is not the intention.

Results

A few results from benchmarks performed on a test machine are included. They are listed below. The command found before each result listing can be used to run the benchmark that generated the result in question.

The test machine is a RaspberryPI Model-B Rev2, running at 700 MHz with the debian style packaged kernel provided by Raspbian. Hopefully it will be easy to replicate the results posted below for anyone interested in doing so.

112 MB; 19 directories, 330 files

Block size 4096 bytes, 10 passes

Command:

linux-microfs/tools/rofsbench.sh -n 10 -r 1387916272 \
    -b 4096 -B 117440512 -w /home/pi/perf/

Result:

Test 0: list all recursively
Command: `ls -lAR /home/pi/perf/tmpfs/mnt`
    microfs:  real=0.2050    sys=0.1030    user=0.0480
    squashfs: real=0.1710    sys=0.0690    user=0.0520
    cramfs:   real=0.1610    sys=0.0480    user=0.0690

Test 1: find all
Command: `find /home/pi/perf/tmpfs/mnt`
    microfs:  real=0.0600    sys=0.0180    user=0.0050
    squashfs: real=0.0600    sys=0.0090    user=0.0110
    cramfs:   real=0.0600    sys=0.0120    user=0.0080

Test 2: find files
Command: `find /home/pi/perf/tmpfs/mnt -type f`
    microfs:  real=0.0700    sys=0.0150    user=0.0150
    squashfs: real=0.0620    sys=0.0190    user=0.0110
    cramfs:   real=0.0630    sys=0.0140    user=0.0140

Test 3: seq access, seq reading
Command: `tools/frd -e -s 1387916272 -i /home/pi/perf/tmpfs/tmp/file-paths.txt`
    microfs:  real=9.0320    sys=6.7580    user=0.0430
    squashfs: real=10.2740   sys=5.7360    user=0.0510
    cramfs:   real=7.7970    sys=5.6260    user=0.0430

Test 4: seq access, stat-only
Command: `tools/frd -e -s 1387916272 -N -i /home/pi/perf/tmpfs/tmp/all-paths.txt`
    microfs:  real=0.0210    sys=0.0070    user=0.0030
    squashfs: real=0.0200    sys=0.0050    user=0.0050
    cramfs:   real=0.0200    sys=0.0060    user=0.0040

Test 5: rand access, seq reading
Command: `tools/frd -e -s 1387916272 -R -i /home/pi/perf/tmpfs/tmp/file-paths.txt`
    microfs:  real=9.0890    sys=6.7960    user=0.0450
    squashfs: real=10.2780   sys=5.7450    user=0.0440
    cramfs:   real=7.8140    sys=5.6560    user=0.0360

Test 6: rand access, stat-only
Command: `tools/frd -e -s 1387916272 -R -N -i /home/pi/perf/tmpfs/tmp/all-paths.txt`
    microfs:  real=0.0200    sys=0.0080    user=0.0020
    squashfs: real=0.0200    sys=0.0070    user=0.0030
    cramfs:   real=0.0200    sys=0.0080    user=0.0020

Test 7: seq access, rand reading
Command: `tools/frd -e -s 1387916272 -r -i /home/pi/perf/tmpfs/tmp/file-paths.txt`
    microfs:  real=9.0670    sys=6.7890    user=0.0720
    squashfs: real=10.6380   sys=6.0060    user=0.0810
    cramfs:   real=8.7300    sys=6.4380    user=0.0990

Test 8: rand access, rand reading
Command: `tools/frd -e -s 1387916272 -R -r -i /home/pi/perf/tmpfs/tmp/file-paths.txt`
    microfs:  real=9.0440    sys=6.7530    user=0.0970
    squashfs: real=10.6460   sys=6.0300    user=0.0690
    cramfs:   real=8.7060    sys=6.4240    user=0.1090

Block size 131072 bytes, 10 passes

Command:

linux-microfs/tools/rofsbench.sh -n 10 -r 1387916272 \
    -b 131072 -B 117440512 -w /home/pi/perf/

Result:

Test 0: list all recursively
Command: `ls -lAR /home/pi/perf/tmpfs/mnt`
    microfs:  real=0.2010    sys=0.1040    user=0.0460
    squashfs: real=0.1630    sys=0.0650    user=0.0550
    cramfs:   real=0.1670    sys=0.0560    user=0.0630

Test 1: find all
Command: `find /home/pi/perf/tmpfs/mnt`
    microfs:  real=0.0600    sys=0.0200    user=0.0050
    squashfs: real=0.0600    sys=0.0120    user=0.0080
    cramfs:   real=0.0600    sys=0.0100    user=0.0100

Test 2: find files
Command: `find /home/pi/perf/tmpfs/mnt -type f`
    microfs:  real=0.0700    sys=0.0190    user=0.0110
    squashfs: real=0.0600    sys=0.0250    user=0.0050
    cramfs:   real=0.0600    sys=0.0170    user=0.0110

Test 3: seq access, seq reading
Command: `tools/frd -e -s 1387916272 -i /home/pi/perf/tmpfs/tmp/file-paths.txt`
    microfs:  real=6.0160    sys=4.3010    user=0.0550
    squashfs: real=8.5080    sys=4.5840    user=0.0500
    cramfs:   real=7.7740    sys=5.6070    user=0.0600

Test 4: seq access, stat-only
Command: `tools/frd -e -s 1387916272 -N -i /home/pi/perf/tmpfs/tmp/all-paths.txt`
    microfs:  real=0.0200    sys=0.0050    user=0.0050
    squashfs: real=0.0220    sys=0.0110    user=0.0000
    cramfs:   real=0.0200    sys=0.0050    user=0.0050

Test 5: rand access, seq reading
Command: `tools/frd -e -s 1387916272 -R -i /home/pi/perf/tmpfs/tmp/file-paths.txt`
    microfs:  real=6.0290    sys=4.3200    user=0.0480
    squashfs: real=8.6120    sys=4.7280    user=0.0360
    cramfs:   real=7.8170    sys=5.6470    user=0.0480

Test 6: rand access, stat-only
Command: `tools/frd -e -s 1387916272 -R -N -i /home/pi/perf/tmpfs/tmp/all-paths.txt`
    microfs:  real=0.0200    sys=0.0050    user=0.0050
    squashfs: real=0.0200    sys=0.0070    user=0.0030
    cramfs:   real=0.0200    sys=0.0070    user=0.0030

Test 7: seq access, rand reading
Command: `tools/frd -e -s 1387916272 -r -i /home/pi/perf/tmpfs/tmp/file-paths.txt`
    microfs:  real=6.0700    sys=4.3420    user=0.0640
    squashfs: real=8.4540    sys=4.5570    user=0.0540
    cramfs:   real=8.7370    sys=6.4860    user=0.0710

Test 8: rand access, rand reading
Command: `tools/frd -e -s 1387916272 -R -r -i /home/pi/perf/tmpfs/tmp/file-paths.txt`
    microfs:  real=6.0660    sys=4.3280    user=0.0720
    squashfs: real=8.6100    sys=4.7060    user=0.0570
    cramfs:   real=8.7230    sys=6.4480    user=0.0950

112 MB; 0 directories, 36 files

Block size 4096 bytes, 10 passes

Command:

linux-microfs/tools/rofsbench.sh -n 10 -r 1389877799 \
    -b 4096 -B 117440512 -w /home/pi/perf/

Result:

Test 0: list all recursively
Command: `ls -lAR /home/pi/perf/tmpfs/mnt`
    microfs:  real=0.1010    sys=0.0260    user=0.0150
    squashfs: real=0.1030    sys=0.0270    user=0.0130
    cramfs:   real=0.0940    sys=0.0270    user=0.0130

Test 1: find all
Command: `find /home/pi/perf/tmpfs/mnt`
    microfs:  real=0.0520    sys=0.0070    user=0.0030
    squashfs: real=0.0500    sys=0.0080    user=0.0020
    cramfs:   real=0.0520    sys=0.0090    user=0.0020

Test 2: find files
Command: `find /home/pi/perf/tmpfs/mnt -type f`
    microfs:  real=0.0500    sys=0.0060    user=0.0050
    squashfs: real=0.0500    sys=0.0090    user=0.0010
    cramfs:   real=0.0500    sys=0.0050    user=0.0050

Test 3: seq access, seq reading
Command: `tools/frd -e -s 1389877799 -i /home/pi/perf/tmpfs/tmp/file-paths.txt`
    microfs:  real=9.0470    sys=6.7440    user=0.0490
    squashfs: real=10.6530   sys=6.0670    user=0.0410
    cramfs:   real=7.7680    sys=5.6060    user=0.0440

Test 4: seq access, stat-only
Command: `tools/frd -e -s 1389877799 -N -i /home/pi/perf/tmpfs/tmp/all-paths.txt`
    microfs:  real=0.0200    sys=0.0080    user=0.0020
    squashfs: real=0.0210    sys=0.0070    user=0.0030
    cramfs:   real=0.0220    sys=0.0090    user=0.0010

Test 5: rand access, seq reading
Command: `tools/frd -e -s 1389877799 -R -i /home/pi/perf/tmpfs/tmp/file-paths.txt`
    microfs:  real=9.0340    sys=6.7370    user=0.0480
    squashfs: real=10.6470   sys=6.0670    user=0.0450
    cramfs:   real=7.7990    sys=5.6060    user=0.0530

Test 6: rand access, stat-only
Command: `tools/frd -e -s 1389877799 -R -N -i /home/pi/perf/tmpfs/tmp/all-paths.txt`
    microfs:  real=0.0200    sys=0.0080    user=0.0020
    squashfs: real=0.0200    sys=0.0050    user=0.0050
    cramfs:   real=0.0210    sys=0.0100    user=0.0000

Test 7: seq access, rand reading
Command: `tools/frd -e -s 1389877799 -r -i /home/pi/perf/tmpfs/tmp/file-paths.txt`
    microfs:  real=9.0660    sys=6.7780    user=0.0750
    squashfs: real=11.0610   sys=6.3760    user=0.0710
    cramfs:   real=8.7260    sys=6.4560    user=0.0780

Test 8: rand access, rand reading
Command: `tools/frd -e -s 1389877799 -R -r -i /home/pi/perf/tmpfs/tmp/file-paths.txt`
    microfs:  real=9.0850    sys=6.7720    user=0.0890
    squashfs: real=11.0750   sys=6.3770    user=0.0790
    cramfs:   real=8.7440    sys=6.4620    user=0.0910

Block size 131072 bytes, 10 passes

Command:

linux-microfs/tools/rofsbench.sh -n 10 -r 1389877799 \
    -b 131072 -B 117440512 -w /home/pi/perf/

Result:

Test 0: list all recursively
Command: `ls -lAR /home/pi/perf/tmpfs/mnt`
    microfs:  real=0.0860    sys=0.0240    user=0.0160
    squashfs: real=0.0800    sys=0.0200    user=0.0190
    cramfs:   real=0.0820    sys=0.0280    user=0.0080

Test 1: find all
Command: `find /home/pi/perf/tmpfs/mnt`
    microfs:  real=0.0500    sys=0.0050    user=0.0050
    squashfs: real=0.0500    sys=0.0040    user=0.0060
    cramfs:   real=0.0500    sys=0.0040    user=0.0060

Test 2: find files
Command: `find /home/pi/perf/tmpfs/mnt -type f`
    microfs:  real=0.0500    sys=0.0030    user=0.0070
    squashfs: real=0.0520    sys=0.0070    user=0.0040
    cramfs:   real=0.0520    sys=0.0060    user=0.0040

Test 3: seq access, seq reading
Command: `tools/frd -e -s 1389877799 -i /home/pi/perf/tmpfs/tmp/file-paths.txt`
    microfs:  real=5.9700    sys=4.2700    user=0.0420
    squashfs: real=8.4100    sys=4.5170    user=0.0400
    cramfs:   real=7.7640    sys=5.5820    user=0.0470

Test 4: seq access, stat-only
Command: `tools/frd -e -s 1389877799 -N -i /home/pi/perf/tmpfs/tmp/all-paths.txt`
    microfs:  real=0.0100    sys=0.0070    user=0.0030
    squashfs: real=0.0100    sys=0.0050    user=0.0050
    cramfs:   real=0.0110    sys=0.0050    user=0.0050

Test 5: rand access, seq reading
Command: `tools/frd -e -s 1389877799 -R -i /home/pi/perf/tmpfs/tmp/file-paths.txt`
    microfs:  real=5.9820    sys=4.2760    user=0.0370
    squashfs: real=8.4210    sys=4.5160    user=0.0470
    cramfs:   real=7.7580    sys=5.5780    user=0.0460

Test 6: rand access, stat-only
Command: `tools/frd -e -s 1389877799 -R -N -i /home/pi/perf/tmpfs/tmp/all-paths.txt`
    microfs:  real=0.0100    sys=0.0070    user=0.0030
    squashfs: real=0.0110    sys=0.0100    user=0.0000
    cramfs:   real=0.0110    sys=0.0070    user=0.0030

Test 7: seq access, rand reading
Command: `tools/frd -e -s 1389877799 -r -i /home/pi/perf/tmpfs/tmp/file-paths.txt`
    microfs:  real=6.0380    sys=4.2920    user=0.0660
    squashfs: real=8.4040    sys=4.4820    user=0.0620
    cramfs:   real=8.7260    sys=6.4260    user=0.1020

Test 8: rand access, rand reading
Command: `tools/frd -e -s 1389877799 -R -r -i /home/pi/perf/tmpfs/tmp/file-paths.txt`
    microfs:  real=6.0140    sys=4.3030    user=0.0500
    squashfs: real=8.3960    sys=4.4840    user=0.0640
    cramfs:   real=8.7240    sys=6.4560    user=0.0790

A small list of mixed TODOs, FIXMEs, WTFs and the sort

The devtable support seems to be quite buggy when tested on real machines (non-virtual).

Name		Name	Last commit message	Last commit date
Latest commit History 123 Commits
extras		extras
tests		tests
tools		tools
vendor		vendor
.gitignore		.gitignore
LICENSE.txt		LICENSE.txt
Makefile		Makefile
README.md		README.md
dev.h		dev.h
devtable.h		devtable.h
hostprog_microfscki.c		hostprog_microfscki.c
hostprog_microfslib.c		hostprog_microfslib.c
hostprog_microfsmki.c		hostprog_microfsmki.c
hostprogs.c		hostprogs.c
hostprogs.h		hostprogs.h
hostprogs_lib.c		hostprogs_lib.c
hostprogs_lib.h		hostprogs_lib.h
hostprogs_lib_lz4.c		hostprogs_lib_lz4.c
hostprogs_lib_lzo.c		hostprogs_lib_lzo.c
hostprogs_lib_xz.c		hostprogs_lib_xz.c
hostprogs_lib_zlib.c		hostprogs_lib_zlib.c
hostprogs_lib_zstd.c		hostprogs_lib_zstd.c
libinfo.h		libinfo.h
libinfo_lz4.h		libinfo_lz4.h
libinfo_lzo.h		libinfo_lzo.h
libinfo_xz.h		libinfo_xz.h
libinfo_zlib.h		libinfo_zlib.h
libinfo_zstd.h		libinfo_zstd.h
microfs.h		microfs.h
microfs_compat.c		microfs_compat.c
microfs_compat.h		microfs_compat.h
microfs_decompressor.c		microfs_decompressor.c
microfs_decompressor_data.c		microfs_decompressor_data.c
microfs_decompressor_data_percpu.c		microfs_decompressor_data_percpu.c
microfs_decompressor_data_queue.c		microfs_decompressor_data_queue.c
microfs_decompressor_data_singleton.c		microfs_decompressor_data_singleton.c
microfs_decompressor_impl_buffer.c		microfs_decompressor_impl_buffer.c
microfs_decompressor_lz4.c		microfs_decompressor_lz4.c
microfs_decompressor_lzo.c		microfs_decompressor_lzo.c
microfs_decompressor_xz.c		microfs_decompressor_xz.c
microfs_decompressor_zlib.c		microfs_decompressor_zlib.c
microfs_decompressor_zstd.c		microfs_decompressor_zstd.c
microfs_fs.h		microfs_fs.h
microfs_inode.c		microfs_inode.c
microfs_read.c		microfs_read.c
microfs_super.c		microfs_super.c
test.c		test.c
test.h		test.h
test.sh		test.sh
test_hostprogs.c		test_hostprogs.c
test_master.c		test_master.c

License

edlund/linux-microfs

Folders and files

Latest commit

History

Repository files navigation

microfs - Minimally Improved Compressed Read Only File System

Licensing

Linux kernel compatibility

Image format

Building microfs

Mount options

Decompressor data

Testing microfs

Reproducible "randomness"

Remote checks

Stress/Load testing

"Quick" tests

Raspberry PI (Raspbian)

Performance

Results

112 MB; 19 directories, 330 files

Block size 4096 bytes, 10 passes

Block size 131072 bytes, 10 passes

112 MB; 0 directories, 36 files

Block size 4096 bytes, 10 passes

Block size 131072 bytes, 10 passes

A small list of mixed TODOs, FIXMEs, WTFs and the sort

About

Topics

Resources

License

Stars

Watchers

Forks

Languages