Merge pull request #9295 from NixOS/store-path-complete-construction

Include store path exact spec in the docs
This commit is contained in:
Robert Hensing 2024-02-12 19:02:03 +01:00 committed by GitHub
commit bdb6f56c90
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
3 changed files with 133 additions and 78 deletions

View file

@ -109,6 +109,7 @@
- [Store Object Info](protocols/json/store-object-info.md)
- [Derivation](protocols/json/derivation.md)
- [Serving Tarball Flakes](protocols/tarball-fetcher.md)
- [Store Path Specification](protocols/store-path.md)
- [Derivation "ATerm" file format](protocols/derivation-aterm.md)
- [Glossary](glossary.md)
- [Contributing](contributing/index.md)

View file

@ -0,0 +1,126 @@
# Complete Store Path Calculation
This is the complete specification for how store paths are calculated.
The format of this specification is close to [Extended BackusNaur form](https://en.wikipedia.org/wiki/Extended_Backus%E2%80%93Naur_form), but must deviate for a few things such as hash functions which we treat as bidirectional for specification purposes.
Regular users do *not* need to know this information --- store paths can be treated as black boxes computed from the properties of the store objects they refer to.
But for those interested in exactly how Nix works, e.g. if they are reimplementing it, this information can be useful.
## Store path proper
```ebnf
store-path = store-dir "/" digest "-" name
```
where
- `name` = the name of the store object.
- `store-dir` = the [store directory](@docroot@/store/store-path.md#store-directory)
- `digest` = base-32 representation of the first 160 bits of a [SHA-256] hash of `fingerprint`
This the hash part of the store name
## Fingerprint
- ```ebnf
fingerprint = type ":" sha256 ":" inner-digest ":" store ":" name
```
Note that it includes the location of the store as well as the name to make sure that changes to either of those are reflected in the hash
(e.g. you won't get `/nix/store/<digest>-name1` and `/nix/store/<digest>-name2`, or `/gnu/store/<digest>-name1`, with equal hash parts).
- `type` = one of:
- ```ebnf
| "text" ( ":" store-path )*
```
for encoded derivations written to the store.
The optional trailing store paths are the references of the store object.
- ```ebnf
| "source" ( ":" store-path )*
```
For paths copied to the store and hashed via a [Nix Archive (NAR)] and [SHA-256][sha-256].
Just like in the text case, we can have the store objects referenced by their paths.
Additionally, we can have an optional `:self` label to denote self reference.
- ```ebnf
| "output:" id
```
For either the outputs built from derivations,
paths copied to the store hashed that area single file hashed directly, or the via a hash algorithm other than [SHA-256][sha-256].
(in that case "source" is used; this is only necessary for compatibility).
`id` is the name of the output (usually, "out").
For content-addressed store objects, `id`, is always "out".
- `inner-digest` = base-16 representation of a SHA-256 hash of `inner-fingerprint`
## Inner fingerprint
- `inner-fingerprint` = one of the following based on `type`:
- if `type` = `"text:" ...`:
the string written to the resulting store path.
- if `type` = `"source:" ...`:
the the hash of the [Nix Archive (NAR)] serialization of the [file system object](@docroot@/store/file-system-object.md) of the store object.
- if `type` = `"output:" id`:
- For input-addressed derivation outputs:
the [ATerm](@docroot@/protocols/derivation-aterm.md) serialization of the derivation modulo fixed output derivations.
- For content-addressed store paths:
```ebnf
"fixed:out:" rec algo ":" hash ":"
```
where
- `rec` = one of:
- ```ebnf
| "r:"
```
hashes of the for [Nix Archive (NAR)] (arbitrary file system object) serialization
- ```ebnf
| ""
```
(empty string) for hashes of the flat (single file) serialization
- ```ebnf
algo = "md5" | "sha1" | "sha256"
```
- `hash` = base-16 representation of the path or flat hash of the contents of the path (or expected contents of the path for fixed-output derivations).
Note that `id` = `"out"`, regardless of the name part of the store path.
Also note that NAR + SHA-256 must not use this case, and instead must use the `type` = `"source:" ...` case.
[Nix Archive (NAR)]: @docroot@/glossary.md#gloss-NAR
[sha-256]: https://en.m.wikipedia.org/wiki/SHA-256
### Historical Note
The `type` = `"source:" ...` and `type` = `"output:out"` grammars technically overlap in purpose,
in that both can represent data hashed by its SHA-256 NAR serialization.
The original reason for this way of computing names was to prevent name collisions (for security).
For instance, the thinking was that it shouldn't be feasible to come up with a derivation whose output path collides with the path for a copied source.
The former would have an `inner-fingerprint` starting with `output:out:`, while the latter would have an `inner-fingerprint` starting with `source:`.
Since `64519cfd657d024ae6e2bb74cb21ad21b886fd2a` (2008), however, it was decided that separating derivation-produced vs manually-hashed content-addressed data like this was not useful.
Now, data that is content-addressed with SHA-256 + NAR-serialization always uses the `source:...` construction, regardless of how it was produced (manually or by derivation).
This allows freely switching between using [fixed-output derivations](@docroot@/glossary.md#gloss-fixed-output-derivation) for fetching, and fetching out-of-band and then manually adding.
It also removes the ambiguity from the grammar.

View file

@ -65,85 +65,13 @@ StorePath Store::followLinksToStorePath(std::string_view path) const
}
/* Store paths have the following form:
/*
The exact specification of store paths is in `protocols/store-path.md`
in the Nix manual. These few functions implement that specification.
<realized-path> = <store>/<h>-<name>
where
<store> = the location of the Nix store, usually /nix/store
<name> = a human readable name for the path, typically obtained
from the name attribute of the derivation, or the name of the
source file from which the store path is created. For derivation
outputs other than the default "out" output, the string "-<id>"
is suffixed to <name>.
<h> = base-32 representation of the first 160 bits of a SHA-256
hash of <s>; the hash part of the store name
<s> = the string "<type>:sha256:<h2>:<store>:<name>";
note that it includes the location of the store as well as the
name to make sure that changes to either of those are reflected
in the hash (e.g. you won't get /nix/store/<h>-name1 and
/nix/store/<h>-name2 with equal hash parts).
<type> = one of:
"text:<r1>:<r2>:...<rN>"
for plain text files written to the store using
addTextToStore(); <r1> ... <rN> are the store paths referenced
by this path, in the form described by <realized-path>
"source:<r1>:<r2>:...:<rN>:self"
for paths copied to the store using addToStore() when recursive
= true and hashAlgo = "sha256". Just like in the text case, we
can have the store paths referenced by the path.
Additionally, we can have an optional :self label to denote self
reference.
"output:<id>"
for either the outputs created by derivations, OR paths copied
to the store using addToStore() with recursive != true or
hashAlgo != "sha256" (in that case "source" is used; it's
silly, but it's done that way for compatibility). <id> is the
name of the output (usually, "out").
<h2> = base-16 representation of a SHA-256 hash of <s2>
<s2> =
if <type> = "text:...":
the string written to the resulting store path
if <type> = "source:...":
the serialisation of the path from which this store path is
copied, as returned by hashPath()
if <type> = "output:<id>":
for non-fixed derivation outputs:
the derivation (see hashDerivationModulo() in
primops.cc)
for paths copied by addToStore() or produced by fixed-output
derivations:
the string "fixed:out:<rec><algo>:<hash>:", where
<rec> = "r:" for recursive (path) hashes, or "" for flat
(file) hashes
<algo> = "md5", "sha1" or "sha256"
<hash> = base-16 representation of the path or flat hash of
the contents of the path (or expected contents of the
path for fixed-output derivations)
Note that since an output derivation has always type output, while
something added by addToStore can have type output or source depending
on the hash, this means that the same input can be hashed differently
if added to the store via addToStore or via a derivation, in the sha256
recursive case.
It would have been nicer to handle fixed-output derivations under
"source", e.g. have something like "source:<rec><algo>", but we're
stuck with this for now...
The main reason for this way of computing names is to prevent name
collisions (for security). For instance, it shouldn't be feasible
to come up with a derivation whose output path collides with the
path for a copied source. The former would have a <s> starting with
"output:out:", while the latter would have a <s> starting with
"source:".
If changes to these functions go beyond mere implementation changes i.e.
also update the user-visible behavior, please update the specification
to match.
*/