1
Fork 0
turbonss/README.md

7.0 KiB

Turbo NSS

Turbonss is a plugin for GNU Name Service Switch (NSS) functionality of GNU C Library (glibc). Turbonss implements lookup for user and passwd database entries (i.e. system users, groups, and group memberships). It's main goal is to run id(1) as fast as possible.

Turbonss is optimized for reading. If the data changes in any way, the whole file will need to be regenerated. Therefore, it was created, and best suited, for environments that have a central user & group database which then needs to be distributed to many servers/services, and the data does not change very often.

This is the fastest known NSS passwd/group implementation for reads. On my 2018-era laptop a corpus with 10k users, 10k groups and 500 average members per group, id takes 17 seconds with the glibc default implementation, 10-17 milliseconds with a pre-cached nscd, ~8 milliseconds with uncached turbonss.

Due to the nature of being built with Zig, this will work on glibc versions as old as 2.16 (may work with even older ones, I did not test beyond that).

Project goals

  • Make it as fast as possible. Especially optimize for the id command.
  • Small database size (helps making it fast).
  • No runtime, no GC, as little as possible overhead.
  • Easy to compile for ancient glibc versions (comes out of the box with Zig).

Dependencies

  1. zig around 0.11.0-dev.3735+a72d634b7.
  2. cmph: bundled with this repository.

Building

Clone, compile and run tests first:

$ git clone --recursive https://git.jakstys.lt/motiejus/turbonss
$ zig build test
$ zig build -Dtarget=x86_64-linux-gnu.2.16 -Doptimize=ReleaseSafe

One may choose different options, depending on requirements. Here are some hints:

  1. -Dcpu=<...> for the CPU microarchitecture.
  2. -Dstrip=true to strip debug symbols.

For reference, size of the shared library and helper binaries when compiled with zig build -Dstrip=true -Doptimize=ReleaseSmall:

$ ls -h1s zig-out/{bin/*,lib/libnss_turbo.so.2.0.0}
 24K zig-out/bin/turbonss-analyze
 20K zig-out/bin/turbonss-getent
 24K zig-out/bin/turbonss-makecorpus
136K zig-out/bin/turbonss-unix2db
 20K zig-out/lib/libnss_turbo.so.2.0.0

Many thanks to Ulrich Drepper for teaching how to link it properly.

Demo

turbonss is best tested, of course, with many users and groups. The guide below will show how to synthesize 10k users, 10k groups with an avereage membership of 1k users per group, and test ubernss with such corpus.

  1. Synthesize some users and groups to passwd and group in the current directory:

    $ zig-out/bin/turbonss-makecorpus
    wrote users=10000 groups=10000 max-members=1000 to .
    $ ls -1hs passwd group
     48M group
    668K passwd
    
  2. Convert the generated passwd and group to the turbonss database. Note the db.turbo database is more than 4 times smaller than the textual one:

    $ zig-out/bin/turbonss-unix2db --group group --passwd passwd
    total 10968064 bytes. groups=10000 users=10000
    $ ls -1hs db.turbo
    11M db.turbo
    
  3. Optional: inspect the freshly created database:

    $ zig-out/bin/turbonss-analyze db.turbo
    File:               db.turbo
    Size:               10,968,064 bytes
    Version:            0
    Endian:             little
    Pointer size:       8 bytes
    getgr buffer size:  18000
    getpw buffer size:  57
    Users:              10000
    Groups:             10000
    Shells:             4
    Most memberships:   u_1000000 (501)
    Sections:
        Name                 Begin    End          Size bytes
        header               00000000 00000080            128
        bdz_gid              00000080 00000e40          3,520
        bdz_groupname        00000e40 00001c00          3,520
        bdz_uid              00001c00 000029c0          3,520
        bdz_username         000029c0 00003780          3,520
        idx_gid2group        00003780 0000d3c0         40,000
        idx_groupname2group  0000d3c0 00017000         40,000
        idx_uid2user         00017000 00020c40         40,000
        idx_name2user        00020c40 0002a880         40,000
        shell_index          0002a880 0002a8c0             64
        shell_blob           0002a8c0 0002a900             64
        groups               0002a900 00065280        240,000
        users                00065280 000da580        480,000
        groupmembers         000da580 005a69c0      5,030,976
        additional_gids      005a69c0 00a75c00      5,042,752
    $ zig-out/bin/turbonss-getent --db db.turbo passwd u_1000000
    u_1000000:x:1000000:1000000:User 1000000:/home/u_1000000:/bin/bash
    $ zig-out/bin/turbonss-getent --db db.turbo group g_1000003
    g_1000003:x:1000003:u_1000002,u_1000003,u_1000004
    
  4. Now since we will be messing with the system, run all following commands in a container:

    $ docker run -ti --rm -v `pwd`:/etc/turbonss -w /etc/turbonss debian:bullseye
    # cp zig-out/lib/libnss_turbo.so.2 /lib/x86_64-linux-gnu/
    
  5. Instruct nsswitch.conf to use both turbonss and the standard resolver:

    # sed -i '/passwd\|group/ s/files/turbo files/' /etc/nsswitch.conf
    # time id u_1000000
    <...>
    real    0m0.006s
    user    0m0.000s
    sys     0m0.008s
    

The id call resolved u_1000000 from db.turbo.

  1. Compare the performance to plain files (that is, without turbonss):

    # sed -i '/passwd\|group/ s/turbo files/files/' /etc/nsswitch.conf
    # cat passwd >> /etc/passwd
    # cat group >> /etc/group
    # time id u_1000000
    <...>
    real    0m17.164s
    user    0m13.288s
    sys     0m3.876s
    

    Over 2500x difference.

More Documentation

  • Architecture is detailed in docs/architecture.md
  • Development notes are in docs/development.md

Project status and known deficiencies

Turbonss works, but, to the author's knowledge, was not deployed to production. If you want to use turbonss instead of a battle-tested, albeit slower nscd, keep the following in mind:

  • turbonss has not been fuzz-tested, so it will crash a program on invalid database file. Please compile with ReleaseSafe. It is plenty fast with this mode, but an invalid database will lead to defined behavior (i.e. crash with a stack trace) instead of overwriting memory wherever.
  • if the database file was replaced while the program has been running, turbonss will not re-read the file (it holds to the previous file descriptor).
  • requires a nightly version of zig (that will change with 0.11).

The license is permissive, so feel free to fork and implement the above (I would appreciate if you told me, but surely you don't have to). I am also available for consulting if that's your preference instead.