1
Fork 0
turbonss/README.md

414 lines
16 KiB
Markdown
Raw Normal View History

2022-02-08 09:52:47 +02:00
Turbo NSS
---------
2022-02-14 10:55:49 +02:00
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
performance, with focus on making [`id(1)`][id] run as fast as possible.
2022-02-14 13:55:54 +02:00
Turbonss is optimized for reading. If the data changes in any way, the whole
file will need to be regenerated (and tooling only supports only full
generation). It was created, and best suited, for environments that have a
central user & group database which then needs to be distributed to many
2022-02-23 10:45:05 +02:00
servers/services, and the data does not change very often (e.g. hourly).
2022-02-14 13:55:54 +02:00
2022-02-14 10:55:49 +02:00
To understand more about name service switch, start with
2022-02-14 13:05:33 +02:00
[`nsswitch.conf(5)`][nsswitch].
2022-02-14 10:55:49 +02:00
Design & constraints
--------------------
2022-02-14 13:37:10 +02:00
To be fast, the user/group database (later: DB) has to be small
([background][data-oriented-design]). It encodes user & group information in a
way that minimizes the DB size, and reduces jumping across the DB ("chasing
pointers and thrashing CPU cache").
2022-02-14 10:55:49 +02:00
2022-02-14 13:37:10 +02:00
To understand how this is done efficiently, let's analyze the
[`getpwnam_r(3)`][getpwnam_r] in high level. This API call accepts a username
and returns the following user information:
2022-02-14 10:55:49 +02:00
```
struct passwd {
char *pw_name; /* username */
char *pw_passwd; /* user password */
uid_t pw_uid; /* user ID */
gid_t pw_gid; /* group ID */
char *pw_gecos; /* user information */
char *pw_dir; /* home directory */
char *pw_shell; /* shell program */
};
```
Turbonss, among others, implements this call, and takes the following steps to
2022-02-14 13:05:33 +02:00
resolve a username to a `struct passwd*`:
2022-02-14 10:55:49 +02:00
2022-02-23 10:45:05 +02:00
- Open the DB (using `mmap`) and interpret it's first 64 bytes as a `*struct
2022-02-14 13:37:10 +02:00
Header`. The header stores offsets to the sections of the file. This needs to
2022-02-23 10:45:05 +02:00
be done once, when the NSS library is loaded.
2022-02-14 10:55:49 +02:00
- Hash the username using a perfect hash function. Perfect hash function
2022-02-14 13:05:33 +02:00
returns a number `n ∈ [0,N-1]`, where N is the total number of users.
2022-02-23 10:45:05 +02:00
- Jump to the `n`'th location in the `idx_name2user` section, which contains
the index `i` to the user's information.
- Jump to the location `i` of section `Users`, which stores the full user
information.
2022-02-14 10:55:49 +02:00
- Decode the user information (which is all in a continuous memory block) and
return it to the caller.
In total, that's one hash for the username (~150ns), two pointer jumps within
2022-02-19 16:09:46 +02:00
the group file (to sections `idx_name2user` and `Users`), and, now that the
2022-02-14 13:37:10 +02:00
user record is found, `memcpy` for each field.
2022-02-14 10:55:49 +02:00
2022-02-23 10:45:05 +02:00
The turbonss DB file is be `mmap`-ed, making it simple to jump across the file
using pointer arithmetic. This also reduces memory usage, as the mmap'ed
regions are shared. Turbonss reads do not consume any heap space.
2022-02-14 13:05:33 +02:00
Tight packing places some constraints on the underlying data:
2022-02-14 10:55:49 +02:00
2022-02-15 10:49:03 +02:00
- Permitted length of username and groupname: 1-32 bytes.
- Permitted length of shell and home: 1-256 bytes.
2022-02-24 05:32:27 +02:00
- Permitted comment ("gecos") length: 0-255 bytes.
- User name, groupname, gecos and shell must be utf8-encoded.
- User and Groups sections are up to 2^35B (~34GB) large. Assuming an "average"
user record takes 50 bytes, this section would fit ~660M users. The
worst-case upper bound is left as an exercise to the reader.
2022-02-08 09:52:47 +02:00
Sorting is stable. In v0:
- Groups are sorted by gid, ascending.
- Users are sorted by their name, ascending by the unicode codepoints
(locale-independent).
Checking out and building
-------------------------
```
$ git clone --recursive https://git.sr.ht/~motiejus/turbonss
```
Alternatively, if you forgot `--recursive`:
```
$ git submodule update --init
```
And run tests:
```
$ zig build test
```
2022-02-13 18:01:44 +02:00
Other commands will be documented as they are implemented.
2022-02-18 17:24:22 +02:00
This project uses [git subtrac][git-subtrac] for managing dependencies. They
work just like regular submodules, except all the refs of the submodules are in
this repository. Repeat after me: all the submodules are in this repository.
So if you have a copy of this repo, dependencies will not disappear.
2022-02-13 18:01:44 +02:00
remarks on `id(1)`
------------------
2022-02-08 09:52:47 +02:00
2022-02-13 18:01:44 +02:00
A known implementation runs id(1) at ~250 rps sequentially on ~20k users and
2022-02-24 05:32:27 +02:00
~10k groups. Our rps target is much higher.
2022-02-08 09:52:47 +02:00
2022-02-14 13:05:33 +02:00
To better reason about the trade-offs, it is useful to understand how `id(1)`
is implemented, in rough terms:
- lookup user by name ([`getpwent_r(3)`][getpwent]).
2022-02-23 10:45:05 +02:00
- get all gids for the user ([`getgrouplist(3)`][getgrouplist]). Note: it is
actually using `initgroups_dyn`, accepts a uid, and is very poorly
documented.
- for each additional gid, get the `struct group*`
([`getgrgid_r(3)`][getgrgid_r]).
2022-02-08 09:52:47 +02:00
2022-02-24 05:32:27 +02:00
Assuming a member is in ~100 groups on average, to reach 10k id/s translates to
1M group lookups per second. We need to convert gid to a group index, and group
index to a group gid/name quickly.
2022-02-08 09:52:47 +02:00
2022-02-13 18:01:44 +02:00
Caveat: `struct group` contains an array of pointers to names of group members
2022-02-14 13:05:33 +02:00
(`char **gr_mem`). However, `id` does not use that information, resulting in
2022-02-23 10:45:05 +02:00
read amplification, sometimes by 10-100x. Therefore, if `argv[0] == "id"`, our
implementation of [`getgrid_r(3)`][getgrid] returns the `struct group*` without
the members. This speeds up `id` by about 10x on a known NSS implementation.
2022-02-13 18:01:44 +02:00
2022-02-23 10:45:05 +02:00
Relatedly, because [`getgrid_r(3)`][getgrid] does not need the group members,
the group members are stored in a different DB section, reducing the `Groups`
section and making more of it fit the CPU caches.
2022-02-11 13:31:54 +02:00
2022-02-13 18:01:44 +02:00
Turbonss header
---------------
2022-02-11 15:37:23 +02:00
2022-02-13 10:42:40 +02:00
The turbonss header looks like this:
2022-02-11 15:37:23 +02:00
```
2022-03-17 07:10:39 +02:00
OFFSET TYPE NAME DESCRIPTION
0 [4]u8 magic f0 9f a4 b7
4 u8 version 0
5 u8 endian 0 for little, 1 for big
2022-03-17 07:10:39 +02:00
6 u8 nblocks_shell_blob max value: 63
7 u8 num_shells max value: 63
8 u32 num_groups number of group entries
12 u32 num_users number of passwd entries
16 u32 nblocks_bdz_gid bdz_gid section block count
20 u32 nblocks_bdz_groupname
24 u32 nblocks_bdz_uid
28 u32 nblocks_bdz_username
32 u64 nblocks_groups
40 u64 nblocks_users
48 u64 nblocks_groupmembers
2022-03-17 07:25:47 +02:00
56 u64 nblocks_additional_gids
2022-04-19 06:11:37 +03:00
64 u64 getgr_bufsize
72 u64 getpw_bufsize
80 [48]u8 padding
2022-02-12 12:30:50 +02:00
```
2022-02-13 18:01:44 +02:00
`magic` is 0xf09fa4b7, and `version` must be `0`. All integers are
2022-03-17 07:10:39 +02:00
native-endian. `nblocks_*` is the count of blocks of a particular section; this
helps calculate the offsets to all sections.
Some numbers, like `nblocks_shell_blob`, `num_shells`, would fit to smaller
number of bytes. However, interpreting `[2]u6` with `xxd(1)` is harder than
interpreting `[2]u8`. Therefore we are using the space we have to make these
integers byte-wide.
2022-02-17 11:16:30 +02:00
2022-04-19 06:11:37 +03:00
`getgr_bufsize` and `getpw_bufsize` is a hint for the caller of `getgr*` and
`getpw*`-family calls. This is the recommended size of the buffer, so the
caller does not receive `ENOMEM`.
2022-02-14 10:55:49 +02:00
Primitive types
---------------
2022-02-12 12:30:50 +02:00
2022-02-23 10:45:05 +02:00
`User` and `Group` entries are sorted by the order they were received in the input
file. All entries are aligned to 8 bytes. All `User` and `Group` entries are
2022-02-14 13:05:33 +02:00
referred by their byte offset in the `Users` and `Groups` section relative to
the beginning of the section.
2022-02-12 12:30:50 +02:00
```
2022-02-24 05:51:04 +02:00
const PackedGroup = packed struct {
2022-02-12 12:30:50 +02:00
gid: u32,
padding: u3,
groupname_len: u5,
2022-02-12 12:30:50 +02:00
}
2022-03-17 07:25:47 +02:00
```
PackedGroup is followed by the group name (of length `groupname_len`), followed
by a varint-compressed offset to the groupmembers section, followed by 8b padding.
2022-02-12 12:30:50 +02:00
2022-03-17 07:25:47 +02:00
PackedUser is a bit more involved:
```
2022-02-23 10:45:05 +02:00
pub const PackedUser = packed struct {
2022-02-12 10:13:10 +02:00
uid: u32,
gid: u32,
shell_len_or_idx: u8,
2022-03-08 20:44:32 +02:00
shell_here: bool,
2022-02-19 11:35:29 +02:00
name_is_a_suffix: bool,
2022-03-08 20:44:32 +02:00
home_len: u6,
2022-02-19 11:35:29 +02:00
name_len: u5,
2022-03-08 20:44:32 +02:00
gecos_len: u11,
2022-02-11 15:37:23 +02:00
}
2022-02-12 10:14:37 +02:00
```
2022-02-11 15:37:23 +02:00
2022-03-17 07:25:47 +02:00
... followed by `userdata: []u8`:
2022-02-19 11:35:29 +02:00
- home.
2022-02-23 10:45:05 +02:00
- name (optional).
2022-02-14 13:05:33 +02:00
- gecos.
- shell (optional).
2022-03-08 20:44:32 +02:00
- `additional_gids_offset`: varint.
2022-02-14 13:05:33 +02:00
2022-03-17 07:25:47 +02:00
First byte of home is stored right after the `gecos_len` field, and its length
is `home_len`. The same logic applies to all the `stringdata` fields: there is
a way to calculate their relative position from the length of the fields before
them.
2022-02-14 13:05:33 +02:00
2022-07-05 06:02:54 +03:00
PackedUser employs two data-oriented compression techniques:
2022-02-14 13:05:33 +02:00
- shells are often shared across different users, see the "Shells" section.
2022-03-17 07:25:47 +02:00
- `name` is frequently a suffix of `home`. For example, `/home/vidmantas` and
`vidmantas`. In this case storing both name and home is wasteful. Therefore
2022-02-23 10:45:05 +02:00
name has two options:
1. `name_is_a_suffix=true`: name is a suffix of the home dir. Then `name`
starts at the `home_len - name_len`'th byte of `home`, and ends at the same
place as `home`.
2. `name_is_a_suffix=false`: name begins one byte after home, and it's length
is `name_len`.
2022-02-14 10:55:49 +02:00
The last field `additional_gids_offset: varint` points to the `additional_gids`
section for this user.
2022-03-08 20:44:32 +02:00
2022-02-14 10:55:49 +02:00
Shells
------
2022-02-23 10:45:05 +02:00
Normally there is a limited number of separate shells even in huge user
databases. A few examples: `/bin/bash`, `/usr/bin/nologin`, `/bin/zsh` among
others. Therefore, "shells" have an optimization: they can be pointed by in the
external list, or, if they are unique to the user, reside among the user's
data.
2022-02-14 10:55:49 +02:00
255 most popular shells (i.e. referred to by at least two User entries) are
2022-02-15 10:49:03 +02:00
stored externally in "Shells" area. The less popular ones are stored with
2022-02-14 10:55:49 +02:00
userdata.
2022-02-23 10:45:05 +02:00
Shells section consists of two sub-sections: the index and the blob. The index
is an array of offsets: the i'th shell starts at `offsets[i]` byte, and ends at
`offsets[i+1]` byte. If there is at least one shell in the shell section, the
index contains a sentinel index as the last element, which signifies the position
of the last byte of the shell blob.
2022-02-15 10:49:03 +02:00
`shell_here=true` in the User struct means the shell is stored with userdata,
and it's length is `shell_len_or_idx`. `shell_here=false` means it is stored in
the `Shells` section, and it's index is `shell_len_or_idx` (and the actual
string start and end offsets are resolved as described in the paragraph above).
2022-02-14 10:55:49 +02:00
2022-02-14 13:05:33 +02:00
Variable-length integers (varints)
----------------------------------
Varint is an efficiently encoded integer (packed for small values). Same as
[protocol buffer varints][varint], except the largest possible value is `u64`.
2022-02-23 10:45:05 +02:00
They compress integers well. Varints are stored for group memberships.
2022-02-14 13:05:33 +02:00
2022-02-22 15:04:59 +02:00
Group memberships
-----------------
There are two group memberships at play:
2022-02-24 05:32:27 +02:00
1. Given a group (gid/name), resolve the members' names (e.g. `getgrgid`).
2. Given a username, resolve user's group gids (for `initgroups(3)`).
2022-02-22 15:04:59 +02:00
2022-02-24 05:32:27 +02:00
When group's memberships are resolved in (1), the same call also requires other
2022-02-22 15:04:59 +02:00
group information: gid and group name. Therefore it makes sense to store a
pointer to the group members in the group information itself. However, the
2022-02-23 10:45:05 +02:00
memberships are not *always* necessary (see remarks about `id(1)`), therefore
the memberships will be stored separately, outside of the groups section.
2022-02-24 05:32:27 +02:00
Similarly, when user's groups are resolved in (2), they are not always necessary
(i.e. not part of `struct user*`), therefore the memberships themselves are
stored out of bound.
2022-04-08 15:05:56 +03:00
`groupmembers` and `additional_gids` store group and user memberships
respectively. Membership IDs are packed — not necessitating random access, thus
suitable for compression.
- `groupmembers` consists of a number X followed by a list of offsets to User
records, because `getgr*` returns pointers to membernames, thus a name has to
be immediately resolvable.
2022-04-08 15:05:56 +03:00
- `additional_gids` is a list of gids, because `initgroups_dyn` (and friends)
returns an array of gids.
Each entry of `groupmembers` and `additional_gids` starts with a varint N,
which is the number of upcoming elements. Then N delta-compressed varints,
which are:
2022-04-08 15:05:56 +03:00
- **additional_gids** a list of gids.
- **groupmembers** byte-offsets to the User records in the `users` section.
2022-02-14 10:55:49 +02:00
2022-02-23 10:45:05 +02:00
Indices
-------
2022-02-23 10:45:05 +02:00
Now that we've sketched the implementation of `id(3)`, it's clearer to
understand which operations need to be fast; in order of importance:
1. lookup gid -> group info (this is on hot path in id) without members.
2. lookup username -> user's groups.
3. lookup uid -> user.
4. lookup groupname -> group.
5. lookup username -> user.
2022-02-13 18:01:44 +02:00
2022-02-23 10:45:05 +02:00
These indices can use perfect hashing like [bdz from cmph][cmph]: a perfect
hash hashes a list of bytes to a sequential list of integers. Perfect hashing
algorithms require some space, and take some time to calculate ("hashing
2022-02-24 05:32:27 +02:00
duration"). I've tested BDZ, which hashes `[][]u8` to a sequential list of
2022-02-23 10:45:05 +02:00
integers (not preserving order) and CHM, preserves order. BDZ accepts an
optional argument `3 <= b <= 10`.
* BDZ algorithm requires (b=3, 900KB, b=7, 338KB, b=10, 306KB) for 1M values.
* Latency to resolve 1M keys: (170ms, 180ms, 230ms, respectively).
* Packed vs non-packed latency differences are not meaningful.
CHM retains order, however, 1M keys weigh 8MB. 10k keys are ~20x larger with
CHM than with BDZ, eliminating the benefit of preserved ordering: we can just
have a separate index.
2022-02-24 05:32:27 +02:00
None of the tested perfect hashing algorithms makes the distinction between
existing (in the initial dictionary) and new keys. In other words, HASH(value)
will be pointing to a number `n ∈ [0,N-1]`, regardless whether the value was in
the initial dictionary. Therefore one must always confirm, after calculating
the hash, that the key matches what's been hashed.
2022-07-05 06:02:54 +03:00
`idx_*` sections are of type `[]u32` and are pointing from `hash(key)` to the
respective `Groups` and `Users` entries (from the beginning of the respective
section). Since User and Group records are 8-byte aligned, the actual offset to
the record is acquired by right-shifting this value by 3 bits.
2022-02-24 05:32:27 +02:00
2022-03-17 07:10:39 +02:00
Database file structure
2022-02-23 10:45:05 +02:00
-----------------------
2022-02-14 13:37:10 +02:00
Each section is padded to 64 bytes.
2022-02-13 18:01:44 +02:00
```
2022-03-05 06:08:01 +02:00
SECTION SIZE DESCRIPTION
header 128 see "Turbonss header" section
2022-03-05 06:08:01 +02:00
bdz_gid ? bdz(gid)
bdz_groupname ? bdz(groupname)
bdz_uid ? bdz(uid)
bdz_username ? bdz(username)
2022-03-13 14:22:49 +02:00
idx_gid2group len(group)*4 bdz->offset Groups
idx_groupname2group len(group)*4 bdz->offset Groups
idx_uid2user len(user)*4 bdz->offset Users
idx_name2user len(user)*4 bdz->offset Users
2022-03-17 07:10:39 +02:00
shell_index len(shells)*2 shell index array
shell_blob <= 65280 shell data blob (max 255*256 bytes)
2022-03-05 06:08:01 +02:00
groups ? packed Group entries (8b padding)
users ? packed User entries (8b padding)
2022-03-17 07:10:39 +02:00
groupmembers ? per-group delta varint memberlist (no padding)
2022-03-17 07:25:47 +02:00
additional_gids ? per-user delta varint gidlist (no padding)
2022-02-13 18:01:44 +02:00
```
Section creation order:
2022-03-17 07:25:47 +02:00
1.`bdz_*`.
1.`shell_index`, `shell_blob`.
1.`additional_gids`.
1.`users` requires `additional_gids` and shell.
1.`groupmembers` requires `users`.
1.`groups` requires `groupmembers`.
1.`idx_*`. requires offsets to `groups` and `users`.
1. ✅ Header.
2022-07-04 07:44:20 +03:00
For v2
------
These are desired for the next DB format:
- Compress strings with fsst.
- Trim first 4 bytes from the cmph headers.
2022-07-04 14:33:46 +03:00
Profiling
---------
Prepare `profile.data`:
```
2022-07-04 23:15:16 +03:00
zig build -Drelease-small=true && \
perf record --call-graph=dwarf \
zig-out/bin/turbo-unix2db --passwd passwd2 --group group2
2022-07-04 14:33:46 +03:00
```
Perf interactive:
```
perf report -i perf.data
```
Flame graph:
```
2022-07-04 23:15:16 +03:00
perf script | inferno-collapse-perf | inferno-flamegraph > profile.svg
2022-07-04 14:33:46 +03:00
```
2022-02-18 17:24:22 +02:00
[git-subtrac]: https://apenwarr.ca/log/20191109
2022-02-11 13:31:54 +02:00
[cmph]: http://cmph.sourceforge.net/
2022-02-14 10:55:49 +02:00
[id]: https://linux.die.net/man/1/id
[nsswitch]: https://linux.die.net/man/5/nsswitch.conf
[data-oriented-design]: https://media.handmade-seattle.com/practical-data-oriented-design/
[getpwnam_r]: https://linux.die.net/man/3/getpwnam_r
2022-02-14 13:05:33 +02:00
[varint]: https://developers.google.com/protocol-buffers/docs/encoding#varints
[getpwent]: https://www.man7.org/linux/man-pages/man3/getpwent_r.3.html
2022-02-23 10:45:05 +02:00
[getgrouplist]: https://www.man7.org/linux/man-pages/man3/getgrouplist.3.html
[getgrid]: https://www.man7.org/linux/man-pages/man3/getgrid_r.3.html