Tarball fetcher: Use the content-addressed Git cache

Backported from the lazy-trees branch.

src/libcmd/common-eval-args.cc

@@ -9,6 +9,7 @@
 #include "store-api.hh"
 #include "command.hh"
 #include "tarball.hh"
+#include "fetch-to-store.hh"
 
 namespace nix {
 
@@ -167,8 +168,9 @@ Bindings * MixEvalArgs::getAutoArgs(EvalState & state)
 SourcePath lookupFileArg(EvalState & state, std::string_view s, const Path * baseDir)
 {
     if (EvalSettings::isPseudoUrl(s)) {
-        auto storePath = fetchers::downloadTarball(
-            state.store, EvalSettings::resolvePseudoUrl(s), "source", false).storePath;
+        auto accessor = fetchers::downloadTarball(
+            EvalSettings::resolvePseudoUrl(s)).accessor;
+        auto storePath = fetchToStore(*state.store, SourcePath(accessor), FetchMode::Copy);
         return state.rootPath(CanonPath(state.store->toRealPath(storePath)));
     }
 

src/libexpr/eval.cc

@@ -2794,10 +2794,11 @@ std::optional<std::string> EvalState::resolveSearchPathPath(const SearchPath::Pa
 
     if (EvalSettings::isPseudoUrl(value)) {
         try {
-            auto storePath = fetchers::downloadTarball(
-                store, EvalSettings::resolvePseudoUrl(value), "source", false).storePath;
+            auto accessor = fetchers::downloadTarball(
+                EvalSettings::resolvePseudoUrl(value)).accessor;
+            auto storePath = fetchToStore(*store, SourcePath(accessor), FetchMode::Copy);
             res = { store->toRealPath(storePath) };
-        } catch (FileTransferError & e) {
+        } catch (Error & e) {
             logWarning({
                 .msg = HintFmt("Nix search path entry '%1%' cannot be downloaded, ignoring", value)
             });

src/libexpr/primops/fetchTree.cc

@@ -9,6 +9,7 @@
 #include "tarball.hh"
 #include "url.hh"
 #include "value-to-json.hh"
+#include "fetch-to-store.hh"
 
 #include <ctime>
 #include <iomanip>
@@ -473,7 +474,7 @@ static void fetch(EvalState & state, const PosIdx pos, Value * * args, Value & v
     //       https://github.com/NixOS/nix/issues/4313
     auto storePath =
         unpack
-        ? fetchers::downloadTarball(state.store, *url, name, (bool) expectedHash).storePath
+        ? fetchToStore(*state.store, fetchers::downloadTarball(*url).accessor, FetchMode::Copy, name)
         : fetchers::downloadFile(state.store, *url, name, (bool) expectedHash).storePath;
 
     if (expectedHash) {

src/libfetchers/git-utils.cc

@@ -466,6 +466,22 @@ struct GitRepoImpl : GitRepo, std::enable_shared_from_this<GitRepoImpl>
         else
             throw Error("Commit signature verification on commit %s failed: %s", rev.gitRev(), output);
     }
+
+    Hash treeHashToNarHash(const Hash & treeHash) override
+    {
+        auto accessor = getAccessor(treeHash, false);
+
+        fetchers::Attrs cacheKey({{"_what", "treeHashToNarHash"}, {"treeHash", treeHash.gitRev()}});
+
+        if (auto res = fetchers::getCache()->lookup(cacheKey))
+            return Hash::parseAny(fetchers::getStrAttr(*res, "narHash"), HashAlgorithm::SHA256);
+
+        auto narHash = accessor->hashPath(CanonPath::root);
+
+        fetchers::getCache()->upsert(cacheKey, fetchers::Attrs({{"narHash", narHash.to_string(HashFormat::SRI, true)}}));
+
+        return narHash;
+    }
 };
 
 ref<GitRepo> GitRepo::openRepo(const std::filesystem::path & path, bool create, bool bare)

src/libfetchers/git-utils.hh

@@ -93,6 +93,12 @@ struct GitRepo
     virtual void verifyCommit(
         const Hash & rev,
         const std::vector<fetchers::PublicKey> & publicKeys) = 0;
+
+    /**
+     * Given a Git tree hash, compute the hash of its NAR
+     * serialisation. This is memoised on-disk.
+     */
+    virtual Hash treeHashToNarHash(const Hash & treeHash) = 0;
 };
 
 ref<GitRepo> getTarballCache();

src/libfetchers/tarball.cc

@@ -9,6 +9,9 @@
 #include "types.hh"
 #include "split.hh"
 #include "posix-source-accessor.hh"
+#include "fs-input-accessor.hh"
+#include "store-api.hh"
+#include "git-utils.hh"
 
 namespace nix::fetchers {
 
@@ -57,10 +60,8 @@ DownloadFileResult downloadFile(
             throw;
     }
 
-    // FIXME: write to temporary file.
     Attrs infoAttrs({
         {"etag", res.etag},
-        {"url", res.effectiveUri},
     });
 
     if (res.immutableUrl)
@@ -91,96 +92,98 @@ DownloadFileResult downloadFile(
         storePath = std::move(info.path);
     }
 
-    getCache()->add(
-        *store,
-        inAttrs,
-        infoAttrs,
-        *storePath,
-        locked);
-
-    if (url != res.effectiveUri)
+    /* Cache metadata for all URLs in the redirect chain. */
+    for (auto & url : res.urls) {
+        inAttrs.insert_or_assign("url", url);
+        infoAttrs.insert_or_assign("url", *res.urls.rbegin());
         getCache()->add(
             *store,
-            {
-                {"type", "file"},
-                {"url", res.effectiveUri},
-                {"name", name},
-            },
+            inAttrs,
             infoAttrs,
             *storePath,
             locked);
+    }
 
     return {
         .storePath = std::move(*storePath),
         .etag = res.etag,
-        .effectiveUrl = res.effectiveUri,
+        .effectiveUrl = *res.urls.rbegin(),
         .immutableUrl = res.immutableUrl,
     };
 }
 
 DownloadTarballResult downloadTarball(
-    ref<Store> store,
     const std::string & url,
-    const std::string & name,
-    bool locked,
     const Headers & headers)
 {
     Attrs inAttrs({
-        {"type", "tarball"},
+        {"_what", "tarballCache"},
         {"url", url},
-        {"name", name},
     });
 
-    auto cached = getCache()->lookupExpired(*store, inAttrs);
+    auto cached = getCache()->lookupExpired(inAttrs);
 
-    if (cached && !cached->expired)
-        return {
-            .storePath = std::move(cached->storePath),
-            .lastModified = (time_t) getIntAttr(cached->infoAttrs, "lastModified"),
-            .immutableUrl = maybeGetStrAttr(cached->infoAttrs, "immutableUrl"),
+    auto attrsToResult = [&](const Attrs & infoAttrs)
+    {
+        auto treeHash = getRevAttr(infoAttrs, "treeHash");
+        return DownloadTarballResult {
+            .treeHash = treeHash,
+            .lastModified = (time_t) getIntAttr(infoAttrs, "lastModified"),
+            .immutableUrl = maybeGetStrAttr(infoAttrs, "immutableUrl"),
+            .accessor = getTarballCache()->getAccessor(treeHash, false),
         };
+    };
 
-    auto res = downloadFile(store, url, name, locked, headers);
+    if (cached && !getTarballCache()->hasObject(getRevAttr(cached->infoAttrs, "treeHash")))
+        cached.reset();
 
-    std::optional<StorePath> unpackedStorePath;
-    time_t lastModified;
+    if (cached && !cached->expired)
+        return attrsToResult(cached->infoAttrs);
 
-    if (cached && res.etag != "" && getStrAttr(cached->infoAttrs, "etag") == res.etag) {
-        unpackedStorePath = std::move(cached->storePath);
-        lastModified = getIntAttr(cached->infoAttrs, "lastModified");
-    } else {
-        Path tmpDir = createTempDir();
-        AutoDelete autoDelete(tmpDir, true);
-        unpackTarfile(store->toRealPath(res.storePath), tmpDir);
-        auto members = readDirectory(tmpDir);
-        if (members.size() != 1)
-            throw nix::Error("tarball '%s' contains an unexpected number of top-level files", url);
-        auto topDir = tmpDir + "/" + members.begin()->name;
-        lastModified = lstat(topDir).st_mtime;
-        PosixSourceAccessor accessor;
-        unpackedStorePath = store->addToStore(name, accessor, CanonPath { topDir }, FileIngestionMethod::Recursive, HashAlgorithm::SHA256, {}, defaultPathFilter, NoRepair);
-    }
+    auto _res = std::make_shared<Sync<FileTransferResult>>();
 
-    Attrs infoAttrs({
-        {"lastModified", uint64_t(lastModified)},
-        {"etag", res.etag},
+    auto source = sinkToSource([&](Sink & sink) {
+        FileTransferRequest req(url);
+        req.expectedETag = cached ? getStrAttr(cached->infoAttrs, "etag") : "";
+        getFileTransfer()->download(std::move(req), sink,
+            [_res](FileTransferResult r)
+            {
+                *_res->lock() = r;
+            });
     });
 
-    if (res.immutableUrl)
-        infoAttrs.emplace("immutableUrl", *res.immutableUrl);
+    // TODO: fall back to cached value if download fails.
 
-    getCache()->add(
-        *store,
-        inAttrs,
-        infoAttrs,
-        *unpackedStorePath,
-        locked);
+    /* Note: if the download is cached, `importTarball()` will receive
+       no data, which causes it to import an empty tarball. */
+    TarArchive archive { *source };
+    auto parseSink = getTarballCache()->getFileSystemObjectSink();
+    auto lastModified = unpackTarfileToSink(archive, *parseSink);
 
-    return {
-        .storePath = std::move(*unpackedStorePath),
-        .lastModified = lastModified,
-        .immutableUrl = res.immutableUrl,
-    };
+    auto res(_res->lock());
+
+    Attrs infoAttrs;
+
+    if (res->cached) {
+        infoAttrs = cached->infoAttrs;
+    } else {
+        infoAttrs.insert_or_assign("etag", res->etag);
+        infoAttrs.insert_or_assign("treeHash", parseSink->sync().gitRev());
+        infoAttrs.insert_or_assign("lastModified", uint64_t(lastModified));
+        if (res->immutableUrl)
+            infoAttrs.insert_or_assign("immutableUrl", *res->immutableUrl);
+    }
+
+    /* Insert a cache entry for every URL in the redirect chain. */
+    for (auto & url : res->urls) {
+        inAttrs.insert_or_assign("url", url);
+        getCache()->upsert(inAttrs, infoAttrs);
+    }
+
+    // FIXME: add a cache entry for immutableUrl? That could allow
+    // cache poisoning.
+
+    return attrsToResult(infoAttrs);
 }
 
 // An input scheme corresponding to a curl-downloadable resource.
@@ -198,6 +201,8 @@ struct CurlInputScheme : InputScheme
 
     virtual bool isValidURL(const ParsedURL & url, bool requireTree) const = 0;
 
+    static const std::set<std::string> specialParams;
+
     std::optional<Input> inputFromURL(const ParsedURL & _url, bool requireTree) const override
     {
         if (!isValidURL(_url, requireTree))
@@ -220,8 +225,12 @@ struct CurlInputScheme : InputScheme
             if (auto n = string2Int<uint64_t>(*i))
                 input.attrs.insert_or_assign("revCount", *n);
 
-        url.query.erase("rev");
-        url.query.erase("revCount");
+        if (auto i = get(url.query, "lastModified"))
+            if (auto n = string2Int<uint64_t>(*i))
+                input.attrs.insert_or_assign("lastModified", *n);
+
+        for (auto & param : allowedAttrs())
+            url.query.erase(param);
 
         input.attrs.insert_or_assign("type", std::string { schemeName() });
         input.attrs.insert_or_assign("url", url.to_string());
@@ -275,10 +284,20 @@ struct FileInputScheme : CurlInputScheme
                 : (!requireTree && !hasTarballExtension(url.path)));
     }
 
-    std::pair<StorePath, Input> fetch(ref<Store> store, const Input & input) override
+    std::pair<ref<InputAccessor>, Input> getAccessor(ref<Store> store, const Input & _input) const override
     {
+        auto input(_input);
+
         auto file = downloadFile(store, getStrAttr(input.attrs, "url"), input.getName(), false);
-        return {std::move(file.storePath), input};
+
+        auto narHash = store->queryPathInfo(file.storePath)->narHash;
+        input.attrs.insert_or_assign("narHash", narHash.to_string(HashFormat::SRI, true));
+
+        auto accessor = makeStorePathAccessor(store, file.storePath);
+
+        accessor->setPathDisplay("«" + input.to_string() + "»");
+
+        return {accessor, input};
     }
 };
 
@@ -296,11 +315,13 @@ struct TarballInputScheme : CurlInputScheme
                 : (requireTree || hasTarballExtension(url.path)));
     }
 
-    std::pair<StorePath, Input> fetch(ref<Store> store, const Input & _input) override
+    std::pair<ref<InputAccessor>, Input> getAccessor(ref<Store> store, const Input & _input) const override
     {
-        Input input(_input);
-        auto url = getStrAttr(input.attrs, "url");
-        auto result = downloadTarball(store, url, input.getName(), false);
+        auto input(_input);
+
+        auto result = downloadTarball(getStrAttr(input.attrs, "url"), {});
+
+        result.accessor->setPathDisplay("«" + input.to_string() + "»");
 
         if (result.immutableUrl) {
             auto immutableInput = Input::fromURL(*result.immutableUrl);
@@ -314,7 +335,10 @@ struct TarballInputScheme : CurlInputScheme
         if (result.lastModified && !input.attrs.contains("lastModified"))
             input.attrs.insert_or_assign("lastModified", uint64_t(result.lastModified));
 
-        return {result.storePath, std::move(input)};
+        input.attrs.insert_or_assign("narHash",
+            getTarballCache()->treeHashToNarHash(result.treeHash).to_string(HashFormat::SRI, true));
+
+        return {result.accessor, input};
     }
 };
 

src/libfetchers/tarball.hh

@@ -2,11 +2,13 @@
 
 #include "types.hh"
 #include "path.hh"
+#include "hash.hh"
 
 #include <optional>
 
 namespace nix {
 class Store;
+struct InputAccessor;
 }
 
 namespace nix::fetchers {
@@ -28,16 +30,18 @@ DownloadFileResult downloadFile(
 
 struct DownloadTarballResult
 {
-    StorePath storePath;
+    Hash treeHash;
     time_t lastModified;
     std::optional<std::string> immutableUrl;
+    ref<InputAccessor> accessor;
 };
 
+/**
+ * Download and import a tarball into the Git cache. The result is the
+ * Git tree hash of the root directory.
+ */
 DownloadTarballResult downloadTarball(
-    ref<Store> store,
     const std::string & url,
-    const std::string & name,
-    bool locked,
     const Headers & headers = {});
 
 }