1
1
Fork 0
mirror of https://github.com/NixOS/nix.git synced 2025-11-20 17:29:36 +01:00

Merge branch 'read-only-local-store' into overlayfs-store

This commit is contained in:
Ben Radford 2023-06-19 16:12:04 +01:00
commit 71f3bad749
No known key found for this signature in database
GPG key ID: 9DF5D4640AB888D5
79 changed files with 1152 additions and 367 deletions

View file

@ -4,7 +4,7 @@
#include "worker.hh"
#include "builtins.hh"
#include "builtins/buildenv.hh"
#include "references.hh"
#include "path-references.hh"
#include "finally.hh"
#include "util.hh"
#include "archive.hh"
@ -2389,18 +2389,21 @@ SingleDrvOutputs LocalDerivationGoal::registerOutputs()
continue;
auto references = *referencesOpt;
auto rewriteOutput = [&]() {
auto rewriteOutput = [&](const StringMap & rewrites) {
/* Apply hash rewriting if necessary. */
if (!outputRewrites.empty()) {
if (!rewrites.empty()) {
debug("rewriting hashes in '%1%'; cross fingers", actualPath);
/* FIXME: this is in-memory. */
StringSink sink;
dumpPath(actualPath, sink);
/* FIXME: Is this actually streaming? */
auto source = sinkToSource([&](Sink & nextSink) {
RewritingSink rsink(rewrites, nextSink);
dumpPath(actualPath, rsink);
rsink.flush();
});
Path tmpPath = actualPath + ".tmp";
restorePath(tmpPath, *source);
deletePath(actualPath);
sink.s = rewriteStrings(sink.s, outputRewrites);
StringSource source(sink.s);
restorePath(actualPath, source);
movePath(tmpPath, actualPath);
/* FIXME: set proper permissions in restorePath() so
we don't have to do another traversal. */
@ -2449,7 +2452,7 @@ SingleDrvOutputs LocalDerivationGoal::registerOutputs()
"since recursive hashing is not enabled (one of outputHashMode={flat,text} is true)",
actualPath);
}
rewriteOutput();
rewriteOutput(outputRewrites);
/* FIXME optimize and deduplicate with addToStore */
std::string oldHashPart { scratchPath->hashPart() };
HashModuloSink caSink { outputHash.hashType, oldHashPart };
@ -2487,16 +2490,14 @@ SingleDrvOutputs LocalDerivationGoal::registerOutputs()
Hash::dummy,
};
if (*scratchPath != newInfo0.path) {
// Also rewrite the output path
auto source = sinkToSource([&](Sink & nextSink) {
RewritingSink rsink2(oldHashPart, std::string(newInfo0.path.hashPart()), nextSink);
dumpPath(actualPath, rsink2);
rsink2.flush();
});
Path tmpPath = actualPath + ".tmp";
restorePath(tmpPath, *source);
deletePath(actualPath);
movePath(tmpPath, actualPath);
// If the path has some self-references, we need to rewrite
// them.
// (note that this doesn't invalidate the ca hash we calculated
// above because it's computed *modulo the self-references*, so
// it already takes this rewrite into account).
rewriteOutput(
StringMap{{oldHashPart,
std::string(newInfo0.path.hashPart())}});
}
HashResult narHashAndSize = hashPath(htSHA256, actualPath);
@ -2518,7 +2519,7 @@ SingleDrvOutputs LocalDerivationGoal::registerOutputs()
outputRewrites.insert_or_assign(
std::string { scratchPath->hashPart() },
std::string { requiredFinalPath.hashPart() });
rewriteOutput();
rewriteOutput(outputRewrites);
auto narHashAndSize = hashPath(htSHA256, actualPath);
ValidPathInfo newInfo0 { requiredFinalPath, narHashAndSize.first };
newInfo0.narSize = narHashAndSize.second;

View file

@ -186,9 +186,9 @@ struct curlFileTransfer : public FileTransfer
size_t realSize = size * nmemb;
std::string line((char *) contents, realSize);
printMsg(lvlVomit, "got header for '%s': %s", request.uri, trim(line));
static std::regex statusLine("HTTP/[^ ]+ +[0-9]+(.*)", std::regex::extended | std::regex::icase);
std::smatch match;
if (std::regex_match(line, match, statusLine)) {
if (std::smatch match; std::regex_match(line, match, statusLine)) {
result.etag = "";
result.data.clear();
result.bodySize = 0;
@ -196,9 +196,11 @@ struct curlFileTransfer : public FileTransfer
acceptRanges = false;
encoding = "";
} else {
auto i = line.find(':');
if (i != std::string::npos) {
std::string name = toLower(trim(line.substr(0, i)));
if (name == "etag") {
result.etag = trim(line.substr(i + 1));
/* Hack to work around a GitHub bug: it sends
@ -212,10 +214,22 @@ struct curlFileTransfer : public FileTransfer
debug("shutting down on 200 HTTP response with expected ETag");
return 0;
}
} else if (name == "content-encoding")
}
else if (name == "content-encoding")
encoding = trim(line.substr(i + 1));
else if (name == "accept-ranges" && toLower(trim(line.substr(i + 1))) == "bytes")
acceptRanges = true;
else if (name == "link" || name == "x-amz-meta-link") {
auto value = trim(line.substr(i + 1));
static std::regex linkRegex("<([^>]*)>; rel=\"immutable\"", std::regex::extended | std::regex::icase);
if (std::smatch match; std::regex_match(value, match, linkRegex))
result.immutableUrl = match.str(1);
else
debug("got invalid link header '%s'", value);
}
}
}
return realSize;
@ -345,7 +359,7 @@ struct curlFileTransfer : public FileTransfer
{
auto httpStatus = getHTTPStatus();
char * effectiveUriCStr;
char * effectiveUriCStr = nullptr;
curl_easy_getinfo(req, CURLINFO_EFFECTIVE_URL, &effectiveUriCStr);
if (effectiveUriCStr)
result.effectiveUri = effectiveUriCStr;

View file

@ -80,6 +80,10 @@ struct FileTransferResult
std::string effectiveUri;
std::string data;
uint64_t bodySize = 0;
/* An "immutable" URL for this resource (i.e. one whose contents
will never change), as returned by the `Link: <url>;
rel="immutable"` header. */
std::optional<std::string> immutableUrl;
};
class Store;

View file

@ -77,7 +77,30 @@ Settings::Settings()
allowedImpureHostPrefixes = tokenizeString<StringSet>("/System/Library /usr/lib /dev /bin/sh");
#endif
buildHook = getSelfExe().value_or("nix") + " __build-remote";
/* Set the build hook location
For builds we perform a self-invocation, so Nix has to be self-aware.
That is, it has to know where it is installed. We don't think it's sentient.
Normally, nix is installed according to `nixBinDir`, which is set at compile time,
but can be overridden. This makes for a great default that works even if this
code is linked as a library into some other program whose main is not aware
that it might need to be a build remote hook.
However, it may not have been installed at all. For example, if it's a static build,
there's a good chance that it has been moved out of its installation directory.
That makes `nixBinDir` useless. Instead, we'll query the OS for the path to the
current executable, using `getSelfExe()`.
As a last resort, we resort to `PATH`. Hopefully we find a `nix` there that's compatible.
If you're porting Nix to a new platform, that might be good enough for a while, but
you'll want to improve `getSelfExe()` to work on your platform.
*/
std::string nixExePath = nixBinDir + "/nix";
if (!pathExists(nixExePath)) {
nixExePath = getSelfExe().value_or("nix");
}
buildHook = nixExePath + " __build-remote";
}
void loadConfFile()

View file

@ -710,20 +710,19 @@ public:
Strings{"https://cache.nixos.org/"},
"substituters",
R"(
A list of [URLs of Nix stores](@docroot@/command-ref/new-cli/nix3-help-stores.md#store-url-format)
to be used as substituters, separated by whitespace.
Substituters are tried based on their Priority value, which each substituter can set
independently. Lower value means higher priority.
The default is `https://cache.nixos.org`, with a Priority of 40.
A list of [URLs of Nix stores](@docroot@/command-ref/new-cli/nix3-help-stores.md#store-url-format) to be used as substituters, separated by whitespace.
A substituter is an additional [store]{@docroot@/glossary.md##gloss-store} from which Nix can obtain [store objects](@docroot@/glossary.md#gloss-store-object) instead of building them.
At least one of the following conditions must be met for Nix to use
a substituter:
Substituters are tried based on their priority value, which each substituter can set independently.
Lower value means higher priority.
The default is `https://cache.nixos.org`, which has a priority of 40.
At least one of the following conditions must be met for Nix to use a substituter:
- the substituter is in the [`trusted-substituters`](#conf-trusted-substituters) list
- the user calling Nix is in the [`trusted-users`](#conf-trusted-users) list
In addition, each store path should be trusted as described
in [`trusted-public-keys`](#conf-trusted-public-keys)
In addition, each store path should be trusted as described in [`trusted-public-keys`](#conf-trusted-public-keys)
)",
{"binary-caches"}};

View file

@ -52,14 +52,15 @@ struct LocalStoreConfig : virtual LocalFSStoreConfig
R"(
Allow this store to be opened when its [database](@docroot@/glossary.md#gloss-nix-database) is on a read-only filesystem.
Normally Nix will attempt to open the store database in read-write mode, even for querying (when write access is not needed).
This causes it to fail if the database is on a read-only filesystem.
Normally Nix will attempt to open the store database in read-write mode, even for querying (when write access is not needed), causing it to fail if the database is on a read-only filesystem.
Enable read-only mode to disable locking and open the SQLite database with the [`immutable` parameter](https://www.sqlite.org/c3ref/open.html) set.
**Warning**
Do not use this unless the filesystem is read-only.
Using it when the filesystem is writable can cause incorrect query results or corruption errors if the database is changed by another process.
> **Warning**
> Do not use this unless the filesystem is read-only.
>
> Using it when the filesystem is writable can cause incorrect query results or corruption errors if the database is changed by another process.
> While the filesystem the database resides on might appear to be read-only, consider whether another user or system might have write access to it.
)"};
const std::string name() override { return "Local Store"; }

View file

@ -0,0 +1,73 @@
#include "path-references.hh"
#include "hash.hh"
#include "util.hh"
#include "archive.hh"
#include <map>
#include <cstdlib>
#include <mutex>
#include <algorithm>
namespace nix {
PathRefScanSink::PathRefScanSink(StringSet && hashes, std::map<std::string, StorePath> && backMap)
: RefScanSink(std::move(hashes))
, backMap(std::move(backMap))
{ }
PathRefScanSink PathRefScanSink::fromPaths(const StorePathSet & refs)
{
StringSet hashes;
std::map<std::string, StorePath> backMap;
for (auto & i : refs) {
std::string hashPart(i.hashPart());
auto inserted = backMap.emplace(hashPart, i).second;
assert(inserted);
hashes.insert(hashPart);
}
return PathRefScanSink(std::move(hashes), std::move(backMap));
}
StorePathSet PathRefScanSink::getResultPaths()
{
/* Map the hashes found back to their store paths. */
StorePathSet found;
for (auto & i : getResult()) {
auto j = backMap.find(i);
assert(j != backMap.end());
found.insert(j->second);
}
return found;
}
std::pair<StorePathSet, HashResult> scanForReferences(
const std::string & path,
const StorePathSet & refs)
{
HashSink hashSink { htSHA256 };
auto found = scanForReferences(hashSink, path, refs);
auto hash = hashSink.finish();
return std::pair<StorePathSet, HashResult>(found, hash);
}
StorePathSet scanForReferences(
Sink & toTee,
const Path & path,
const StorePathSet & refs)
{
PathRefScanSink refsSink = PathRefScanSink::fromPaths(refs);
TeeSink sink { refsSink, toTee };
/* Look for the hashes in the NAR dump of the path. */
dumpPath(path, sink);
return refsSink.getResultPaths();
}
}

View file

@ -0,0 +1,25 @@
#pragma once
#include "references.hh"
#include "path.hh"
namespace nix {
std::pair<StorePathSet, HashResult> scanForReferences(const Path & path, const StorePathSet & refs);
StorePathSet scanForReferences(Sink & toTee, const Path & path, const StorePathSet & refs);
class PathRefScanSink : public RefScanSink
{
std::map<std::string, StorePath> backMap;
PathRefScanSink(StringSet && hashes, std::map<std::string, StorePath> && backMap);
public:
static PathRefScanSink fromPaths(const StorePathSet & refs);
StorePathSet getResultPaths();
};
}

View file

@ -13,8 +13,10 @@
namespace nix {
/* Parse a generation name of the format
`<profilename>-<number>-link'. */
/**
* Parse a generation name of the format
* `<profilename>-<number>-link'.
*/
static std::optional<GenerationNumber> parseName(const std::string & profileName, const std::string & name)
{
if (name.substr(0, profileName.size() + 1) != profileName + "-") return {};
@ -28,7 +30,6 @@ static std::optional<GenerationNumber> parseName(const std::string & profileName
}
std::pair<Generations, std::optional<GenerationNumber>> findGenerations(Path profile)
{
Generations gens;
@ -61,15 +62,16 @@ std::pair<Generations, std::optional<GenerationNumber>> findGenerations(Path pro
}
static void makeName(const Path & profile, GenerationNumber num,
Path & outLink)
/**
* Create a generation name that can be parsed by `parseName()`.
*/
static Path makeName(const Path & profile, GenerationNumber num)
{
Path prefix = fmt("%1%-%2%", profile, num);
outLink = prefix + "-link";
return fmt("%s-%s-link", profile, num);
}
Path createGeneration(ref<LocalFSStore> store, Path profile, StorePath outPath)
Path createGeneration(LocalFSStore & store, Path profile, StorePath outPath)
{
/* The new generation number should be higher than old the
previous ones. */
@ -79,7 +81,7 @@ Path createGeneration(ref<LocalFSStore> store, Path profile, StorePath outPath)
if (gens.size() > 0) {
Generation last = gens.back();
if (readLink(last.path) == store->printStorePath(outPath)) {
if (readLink(last.path) == store.printStorePath(outPath)) {
/* We only create a new generation symlink if it differs
from the last one.
@ -89,7 +91,7 @@ Path createGeneration(ref<LocalFSStore> store, Path profile, StorePath outPath)
return last.path;
}
num = gens.back().number;
num = last.number;
} else {
num = 0;
}
@ -100,9 +102,8 @@ Path createGeneration(ref<LocalFSStore> store, Path profile, StorePath outPath)
to the permanent roots (of which the GC would have a stale
view). If we didn't do it this way, the GC might remove the
user environment etc. we've just built. */
Path generation;
makeName(profile, num + 1, generation);
store->addPermRoot(outPath, generation);
Path generation = makeName(profile, num + 1);
store.addPermRoot(outPath, generation);
return generation;
}
@ -117,12 +118,19 @@ static void removeFile(const Path & path)
void deleteGeneration(const Path & profile, GenerationNumber gen)
{
Path generation;
makeName(profile, gen, generation);
Path generation = makeName(profile, gen);
removeFile(generation);
}
/**
* Delete a generation with dry-run mode.
*
* Like `deleteGeneration()` but:
*
* - We log what we are going to do.
*
* - We only actually delete if `dryRun` is false.
*/
static void deleteGeneration2(const Path & profile, GenerationNumber gen, bool dryRun)
{
if (dryRun)
@ -150,27 +158,36 @@ void deleteGenerations(const Path & profile, const std::set<GenerationNumber> &
}
}
/**
* Advanced the iterator until the given predicate `cond` returns `true`.
*/
static inline void iterDropUntil(Generations & gens, auto && i, auto && cond)
{
for (; i != gens.rend() && !cond(*i); ++i);
}
void deleteGenerationsGreaterThan(const Path & profile, GenerationNumber max, bool dryRun)
{
if (max == 0)
throw Error("Must keep at least one generation, otherwise the current one would be deleted");
PathLocks lock;
lockProfile(lock, profile);
bool fromCurGen = false;
auto [gens, curGen] = findGenerations(profile);
for (auto i = gens.rbegin(); i != gens.rend(); ++i) {
if (i->number == curGen) {
fromCurGen = true;
max--;
continue;
}
if (fromCurGen) {
if (max) {
max--;
continue;
}
deleteGeneration2(profile, i->number, dryRun);
}
}
auto [gens, _curGen] = findGenerations(profile);
auto curGen = _curGen;
auto i = gens.rbegin();
// Find the current generation
iterDropUntil(gens, i, [&](auto & g) { return g.number == curGen; });
// Skip over `max` generations, preserving them
for (auto keep = 0; i != gens.rend() && keep < max; ++i, ++keep);
// Delete the rest
for (; i != gens.rend(); ++i)
deleteGeneration2(profile, i->number, dryRun);
}
void deleteOldGenerations(const Path & profile, bool dryRun)
@ -193,23 +210,33 @@ void deleteGenerationsOlderThan(const Path & profile, time_t t, bool dryRun)
auto [gens, curGen] = findGenerations(profile);
bool canDelete = false;
for (auto i = gens.rbegin(); i != gens.rend(); ++i)
if (canDelete) {
assert(i->creationTime < t);
if (i->number != curGen)
deleteGeneration2(profile, i->number, dryRun);
} else if (i->creationTime < t) {
/* We may now start deleting generations, but we don't
delete this generation yet, because this generation was
still the one that was active at the requested point in
time. */
canDelete = true;
}
auto i = gens.rbegin();
// Predicate that the generation is older than the given time.
auto older = [&](auto & g) { return g.creationTime < t; };
// Find the first older generation, if one exists
iterDropUntil(gens, i, older);
/* Take the previous generation
We don't want delete this one yet because it
existed at the requested point in time, and
we want to be able to roll back to it. */
if (i != gens.rend()) ++i;
// Delete all previous generations (unless current).
for (; i != gens.rend(); ++i) {
/* Creating date and generations should be monotonic, so lower
numbered derivations should also be older. */
assert(older(*i));
if (i->number != curGen)
deleteGeneration2(profile, i->number, dryRun);
}
}
void deleteGenerationsOlderThan(const Path & profile, std::string_view timeSpec, bool dryRun)
time_t parseOlderThanTimeSpec(std::string_view timeSpec)
{
if (timeSpec.empty() || timeSpec[timeSpec.size() - 1] != 'd')
throw UsageError("invalid number of days specifier '%1%', expected something like '14d'", timeSpec);
@ -221,9 +248,7 @@ void deleteGenerationsOlderThan(const Path & profile, std::string_view timeSpec,
if (!days || *days < 1)
throw UsageError("invalid number of days specifier '%1%'", timeSpec);
time_t oldTime = curTime - *days * 24 * 3600;
deleteGenerationsOlderThan(profile, oldTime, dryRun);
return curTime - *days * 24 * 3600;
}

View file

@ -1,7 +1,11 @@
#pragma once
///@file
/**
* @file Implementation of Profiles.
*
* See the manual for additional information.
*/
#include "types.hh"
#include "types.hh"
#include "pathlocks.hh"
#include <time.h>
@ -12,41 +16,166 @@ namespace nix {
class StorePath;
/**
* A positive number identifying a generation for a given profile.
*
* Generation numbers are assigned sequentially. Each new generation is
* assigned 1 + the current highest generation number.
*/
typedef uint64_t GenerationNumber;
/**
* A generation is a revision of a profile.
*
* Each generation is a mapping (key-value pair) from an identifier
* (`number`) to a store object (specified by `path`).
*/
struct Generation
{
/**
* The number of a generation is its unique identifier within the
* profile.
*/
GenerationNumber number;
/**
* The store path identifies the store object that is the contents
* of the generation.
*
* These store paths / objects are not unique to the generation
* within a profile. Nix tries to ensure successive generations have
* distinct contents to avoid bloat, but nothing stops two
* non-adjacent generations from having the same contents.
*
* @todo Use `StorePath` instead of `Path`?
*/
Path path;
/**
* When the generation was created. This is extra metadata about the
* generation used to make garbage collecting old generations more
* convenient.
*/
time_t creationTime;
};
/**
* All the generations of a profile
*/
typedef std::list<Generation> Generations;
/**
* Returns the list of currently present generations for the specified
* profile, sorted by generation number. Also returns the number of
* the current generation.
* Find all generations for the given profile.
*
* @param profile A profile specified by its name and location combined
* into a path. E.g. if "foo" is the name of the profile, and "/bar/baz"
* is the directory it is in, then the path "/bar/baz/foo" would be the
* argument for this parameter.
*
* @return The pair of:
*
* - The list of currently present generations for the specified profile,
* sorted by ascending generation number.
*
* - The number of the current/active generation.
*
* Note that the current/active generation need not be the latest one.
*/
std::pair<Generations, std::optional<GenerationNumber>> findGenerations(Path profile);
class LocalFSStore;
Path createGeneration(ref<LocalFSStore> store, Path profile, StorePath outPath);
/**
* Create a new generation of the given profile
*
* If the previous generation (not the currently active one!) has a
* distinct store object, a fresh generation number is mapped to the
* given store object, referenced by path. Otherwise, the previous
* generation is assumed.
*
* The behavior of reusing existing generations like this makes this
* procedure idempotent. It also avoids clutter.
*/
Path createGeneration(LocalFSStore & store, Path profile, StorePath outPath);
/**
* Unconditionally delete a generation
*
* @param profile A profile specified by its name and location combined into a path.
*
* @param gen The generation number specifying exactly which generation
* to delete.
*
* Because there is no check of whether the generation to delete is
* active, this is somewhat unsafe.
*
* @todo Should we expose this at all?
*/
void deleteGeneration(const Path & profile, GenerationNumber gen);
/**
* Delete the given set of generations.
*
* @param profile The profile, specified by its name and location combined into a path, whose generations we want to delete.
*
* @param gensToDelete The generations to delete, specified by a set of
* numbers.
*
* @param dryRun Log what would be deleted instead of actually doing
* so.
*
* Trying to delete the currently active generation will fail, and cause
* no generations to be deleted.
*/
void deleteGenerations(const Path & profile, const std::set<GenerationNumber> & gensToDelete, bool dryRun);
/**
* Delete generations older than `max` passed the current generation.
*
* @param profile The profile, specified by its name and location combined into a path, whose generations we want to delete.
*
* @param max How many generations to keep up to the current one. Must
* be at least 1 so we don't delete the current one.
*
* @param dryRun Log what would be deleted instead of actually doing
* so.
*/
void deleteGenerationsGreaterThan(const Path & profile, GenerationNumber max, bool dryRun);
/**
* Delete all generations other than the current one
*
* @param profile The profile, specified by its name and location combined into a path, whose generations we want to delete.
*
* @param dryRun Log what would be deleted instead of actually doing
* so.
*/
void deleteOldGenerations(const Path & profile, bool dryRun);
/**
* Delete generations older than `t`, except for the most recent one
* older than `t`.
*
* @param profile The profile, specified by its name and location combined into a path, whose generations we want to delete.
*
* @param dryRun Log what would be deleted instead of actually doing
* so.
*/
void deleteGenerationsOlderThan(const Path & profile, time_t t, bool dryRun);
void deleteGenerationsOlderThan(const Path & profile, std::string_view timeSpec, bool dryRun);
/**
* Parse a temp spec intended for `deleteGenerationsOlderThan()`.
*
* Throws an exception if `timeSpec` fails to parse.
*/
time_t parseOlderThanTimeSpec(std::string_view timeSpec);
/**
* Smaller wrapper around `replaceSymlink` for replacing the current
* generation of a profile. Does not enforce proper structure.
*
* @todo Always use `switchGeneration()` instead, and delete this.
*/
void switchLink(Path link, Path target);
/**

View file

@ -1,188 +0,0 @@
#include "references.hh"
#include "hash.hh"
#include "util.hh"
#include "archive.hh"
#include <map>
#include <cstdlib>
#include <mutex>
namespace nix {
static size_t refLength = 32; /* characters */
static void search(
std::string_view s,
StringSet & hashes,
StringSet & seen)
{
static std::once_flag initialised;
static bool isBase32[256];
std::call_once(initialised, [](){
for (unsigned int i = 0; i < 256; ++i) isBase32[i] = false;
for (unsigned int i = 0; i < base32Chars.size(); ++i)
isBase32[(unsigned char) base32Chars[i]] = true;
});
for (size_t i = 0; i + refLength <= s.size(); ) {
int j;
bool match = true;
for (j = refLength - 1; j >= 0; --j)
if (!isBase32[(unsigned char) s[i + j]]) {
i += j + 1;
match = false;
break;
}
if (!match) continue;
std::string ref(s.substr(i, refLength));
if (hashes.erase(ref)) {
debug("found reference to '%1%' at offset '%2%'", ref, i);
seen.insert(ref);
}
++i;
}
}
void RefScanSink::operator () (std::string_view data)
{
/* It's possible that a reference spans the previous and current
fragment, so search in the concatenation of the tail of the
previous fragment and the start of the current fragment. */
auto s = tail;
auto tailLen = std::min(data.size(), refLength);
s.append(data.data(), tailLen);
search(s, hashes, seen);
search(data, hashes, seen);
auto rest = refLength - tailLen;
if (rest < tail.size())
tail = tail.substr(tail.size() - rest);
tail.append(data.data() + data.size() - tailLen, tailLen);
}
PathRefScanSink::PathRefScanSink(StringSet && hashes, std::map<std::string, StorePath> && backMap)
: RefScanSink(std::move(hashes))
, backMap(std::move(backMap))
{ }
PathRefScanSink PathRefScanSink::fromPaths(const StorePathSet & refs)
{
StringSet hashes;
std::map<std::string, StorePath> backMap;
for (auto & i : refs) {
std::string hashPart(i.hashPart());
auto inserted = backMap.emplace(hashPart, i).second;
assert(inserted);
hashes.insert(hashPart);
}
return PathRefScanSink(std::move(hashes), std::move(backMap));
}
StorePathSet PathRefScanSink::getResultPaths()
{
/* Map the hashes found back to their store paths. */
StorePathSet found;
for (auto & i : getResult()) {
auto j = backMap.find(i);
assert(j != backMap.end());
found.insert(j->second);
}
return found;
}
std::pair<StorePathSet, HashResult> scanForReferences(
const std::string & path,
const StorePathSet & refs)
{
HashSink hashSink { htSHA256 };
auto found = scanForReferences(hashSink, path, refs);
auto hash = hashSink.finish();
return std::pair<StorePathSet, HashResult>(found, hash);
}
StorePathSet scanForReferences(
Sink & toTee,
const Path & path,
const StorePathSet & refs)
{
PathRefScanSink refsSink = PathRefScanSink::fromPaths(refs);
TeeSink sink { refsSink, toTee };
/* Look for the hashes in the NAR dump of the path. */
dumpPath(path, sink);
return refsSink.getResultPaths();
}
RewritingSink::RewritingSink(const std::string & from, const std::string & to, Sink & nextSink)
: from(from), to(to), nextSink(nextSink)
{
assert(from.size() == to.size());
}
void RewritingSink::operator () (std::string_view data)
{
std::string s(prev);
s.append(data);
size_t j = 0;
while ((j = s.find(from, j)) != std::string::npos) {
matches.push_back(pos + j);
s.replace(j, from.size(), to);
}
prev = s.size() < from.size() ? s : std::string(s, s.size() - from.size() + 1, from.size() - 1);
auto consumed = s.size() - prev.size();
pos += consumed;
if (consumed) nextSink(s.substr(0, consumed));
}
void RewritingSink::flush()
{
if (prev.empty()) return;
pos += prev.size();
nextSink(prev);
prev.clear();
}
HashModuloSink::HashModuloSink(HashType ht, const std::string & modulus)
: hashSink(ht)
, rewritingSink(modulus, std::string(modulus.size(), 0), hashSink)
{
}
void HashModuloSink::operator () (std::string_view data)
{
rewritingSink(data);
}
HashResult HashModuloSink::finish()
{
rewritingSink.flush();
/* Hash the positions of the self-references. This ensures that a
NAR with self-references and a NAR with some of the
self-references already zeroed out do not produce a hash
collision. FIXME: proof. */
for (auto & pos : rewritingSink.matches)
hashSink(fmt("|%d", pos));
auto h = hashSink.finish();
return {h.first, rewritingSink.pos};
}
}

View file

@ -1,71 +0,0 @@
#pragma once
///@file
#include "hash.hh"
#include "path.hh"
namespace nix {
std::pair<StorePathSet, HashResult> scanForReferences(const Path & path, const StorePathSet & refs);
StorePathSet scanForReferences(Sink & toTee, const Path & path, const StorePathSet & refs);
class RefScanSink : public Sink
{
StringSet hashes;
StringSet seen;
std::string tail;
public:
RefScanSink(StringSet && hashes) : hashes(hashes)
{ }
StringSet & getResult()
{ return seen; }
void operator () (std::string_view data) override;
};
class PathRefScanSink : public RefScanSink
{
std::map<std::string, StorePath> backMap;
PathRefScanSink(StringSet && hashes, std::map<std::string, StorePath> && backMap);
public:
static PathRefScanSink fromPaths(const StorePathSet & refs);
StorePathSet getResultPaths();
};
struct RewritingSink : Sink
{
std::string from, to, prev;
Sink & nextSink;
uint64_t pos = 0;
std::vector<uint64_t> matches;
RewritingSink(const std::string & from, const std::string & to, Sink & nextSink);
void operator () (std::string_view data) override;
void flush();
};
struct HashModuloSink : AbstractHashSink
{
HashSink hashSink;
RewritingSink rewritingSink;
HashModuloSink(HashType ht, const std::string & modulus);
void operator () (std::string_view data) override;
HashResult finish() override;
};
}

View file

@ -29,7 +29,7 @@ enum class SQLiteOpenMode {
* Use this mode if the database is on a read-only filesystem.
* Fails with an error if the database does not exist.
*/
Immutable
Immutable,
};
/**