Implement caching of fetchurl/fetchTarball results

ETags are used to prevent redownloading unchanged files.

4 files changed, 147 insertions, 24 deletions

diff --git a/src/libexpr/primops.cc b/src/libexpr/primops.cc
index e81849646..77ca42042 100644
--- a/src/libexpr/primops.cc
+++ b/src/libexpr/primops.cc
@@ -1490,6 +1490,9 @@ struct Curl
 {
     CURL * curl;
     string data;
+    string etag, status, expectedETag;
+
+    struct curl_slist * requestHeaders;
 
     static size_t writeCallback(void * contents, size_t size, size_t nmemb, void * userp)
     {
@@ -1499,36 +1502,92 @@ struct Curl
         return realSize;
     }
 
+    static size_t headerCallback(void * contents, size_t size, size_t nmemb, void * userp)
+    {
+        Curl & c(* (Curl *) userp);
+        size_t realSize = size * nmemb;
+        string line = string((char *) contents, realSize);
+        printMsg(lvlVomit, format("got header: %1%") % trim(line));
+        if (line.compare(0, 5, "HTTP/") == 0) { // new response starts
+            c.etag = "";
+            auto ss = tokenizeString<vector<string>>(line, " ");
+            c.status = ss.size() >= 2 ? ss[1] : "";
+        } else {
+            auto i = line.find(':');
+            if (i != string::npos) {
+                string name = trim(string(line, 0, i));
+                if (name == "ETag") { // FIXME: case
+                    c.etag = trim(string(line, i + 1));
+                    /* Hack to work around a GitHub bug: it sends
+                       ETags, but ignores If-None-Match. So if we get
+                       the expected ETag on a 200 response, then shut
+                       down the connection because we already have the
+                       data. */
+                    printMsg(lvlDebug, format("got ETag: %1%") % c.etag);
+                    if (c.etag == c.expectedETag && c.status == "200") {
+                        printMsg(lvlDebug, format("shutting down on 200 HTTP response with expected ETag"));
+                        return 0;
+                    }
+                }
+            }
+        }
+        return realSize;
+    }
+
     Curl()
     {
+        requestHeaders = 0;
+
         curl = curl_easy_init();
         if (!curl) throw Error("unable to initialize curl");
 
         curl_easy_setopt(curl, CURLOPT_FOLLOWLOCATION, 1L);
         curl_easy_setopt(curl, CURLOPT_CAINFO, getEnv("SSL_CERT_FILE", "/etc/ssl/certs/ca-certificates.crt").c_str());
         curl_easy_setopt(curl, CURLOPT_USERAGENT, ("Nix/" + nixVersion).c_str());
+        curl_easy_setopt(curl, CURLOPT_FAILONERROR, 1);
 
         curl_easy_setopt(curl, CURLOPT_WRITEFUNCTION, writeCallback);
         curl_easy_setopt(curl, CURLOPT_WRITEDATA, (void *) &curl);
+
+        curl_easy_setopt(curl, CURLOPT_HEADERFUNCTION, headerCallback);
+        curl_easy_setopt(curl, CURLOPT_HEADERDATA, (void *) &curl);
     }
 
     ~Curl()
     {
         if (curl) curl_easy_cleanup(curl);
+        if (requestHeaders) curl_slist_free_all(requestHeaders);
     }
 
-    string fetch(const string & url)
+    bool fetch(const string & url, const string & expectedETag = "")
     {
         curl_easy_setopt(curl, CURLOPT_URL, url.c_str());
 
         data.clear();
 
+        if (requestHeaders) {
+            curl_slist_free_all(requestHeaders);
+            requestHeaders = 0;
+        }
+
+        if (!expectedETag.empty()) {
+            this->expectedETag = expectedETag;
+            requestHeaders = curl_slist_append(requestHeaders, ("If-None-Match: " + expectedETag).c_str());
+        }
+
+        curl_easy_setopt(curl, CURLOPT_HTTPHEADER, requestHeaders);
+
         CURLcode res = curl_easy_perform(curl);
+        if (res == CURLE_WRITE_ERROR && etag == expectedETag) return false;
         if (res != CURLE_OK)
-            throw Error(format("unable to download ‘%1%’: %2%")
-                % url % curl_easy_strerror(res));
+            throw Error(format("unable to download ‘%1%’: %2% (%3%)")
+                % url % curl_easy_strerror(res) % res);
 
-        return data;
+        long httpStatus = 0;
+        curl_easy_getinfo(curl, CURLINFO_RESPONSE_CODE, &httpStatus);
+        if (httpStatus == 304) return false;
+
+        return true;
     }
 };
 
@@ -1560,25 +1619,71 @@ void fetch(EvalState & state, const Pos & pos, Value * * args, Value & v,
     } else
         url = state.forceStringNoCtx(*args[0], pos);
 
-    // TODO: cache downloads.
+    Path cacheDir = getEnv("XDG_CACHE_HOME", getEnv("HOME", "") + "/.cache") + "/nix/tarballs";
+    createDirs(cacheDir);
 
-    Curl curl;
-    string data = curl.fetch(url);
+    string urlHash = printHash32(hashString(htSHA256, url));
+
+    Path dataFile = cacheDir + "/" + urlHash + ".info";
+    Path fileLink = cacheDir + "/" + urlHash + "-file";
+
+    Path storePath;
+
+    string expectedETag;
+
+    if (pathExists(fileLink) && pathExists(dataFile)) {
+        storePath = readLink(fileLink);
+        store->addTempRoot(storePath);
+        if (store->isValidPath(storePath)) {
+            auto ss = tokenizeString<vector<string>>(readFile(dataFile), "\n");
+            if (ss.size() >= 2 && ss[0] == url) {
+                printMsg(lvlDebug, format("verifying previous ETag ‘%1%’") % ss[1]);
+                expectedETag = ss[1];
+            }
+        } else
+            storePath = "";
+    }
 
     string name;
-    string::size_type p = url.rfind('/');
+    auto p = url.rfind('/');
     if (p != string::npos) name = string(url, p + 1);
 
-    Path storePath = store->addTextToStore(name, data, PathSet(), state.repair);
+    if (expectedETag.empty())
+        printMsg(lvlInfo, format("downloading ‘%1%’...") % url);
+    else
+        printMsg(lvlInfo, format("checking ‘%1%’...") % url);
+    Curl curl;
+
+    if (curl.fetch(url, expectedETag))
+        storePath = store->addTextToStore(name, curl.data, PathSet(), state.repair);
+
+    assert(!storePath.empty());
+    replaceSymlink(storePath, fileLink);
+
+    writeFile(dataFile, url + "\n" + curl.etag + "\n");
 
     if (unpack) {
-        Path tmpDir = createTempDir();
-        AutoDelete autoDelete(tmpDir, true);
-        runProgram("tar", true, {"xf", storePath, "-C", tmpDir, "--strip-components", "1"}, "");
-        storePath = store->addToStore(name, tmpDir, true, htSHA256, defaultPathFilter, state.repair);
+        Path unpackedLink = cacheDir + "/" + baseNameOf(storePath) + "-unpacked";
+        Path unpackedStorePath;
+        if (pathExists(unpackedLink)) {
+            unpackedStorePath = readLink(unpackedLink);
+            store->addTempRoot(unpackedStorePath);
+            if (!store->isValidPath(unpackedStorePath))
+                unpackedStorePath = "";
+        }
+        if (unpackedStorePath.empty()) {
+            printMsg(lvlDebug, format("unpacking ‘%1%’...") % storePath);
+            Path tmpDir = createTempDir();
+            AutoDelete autoDelete(tmpDir, true);
+            runProgram("tar", true, {"xf", storePath, "-C", tmpDir, "--strip-components", "1"}, "");
+            unpackedStorePath = store->addToStore(name, tmpDir, true, htSHA256, defaultPathFilter, state.repair);
+        }
+        replaceSymlink(unpackedStorePath, unpackedLink);
+        mkString(v, unpackedStorePath, singleton<PathSet>(unpackedStorePath));
     }
 
-    mkString(v, storePath, singleton<PathSet>(storePath));
+    else
+        mkString(v, storePath, singleton<PathSet>(storePath));
 }
 
 
diff --git a/src/libutil/util.cc b/src/libutil/util.cc
index be0a9bf31..ab0a3b303 100644
--- a/src/libutil/util.cc
+++ b/src/libutil/util.cc
@@ -413,6 +413,17 @@ void createSymlink(const Path & target, const Path & link)
 }
 
 
+void replaceSymlink(const Path & target, const Path & link)
+{
+    Path tmp = canonPath(dirOf(link) + "/.new_" + baseNameOf(link));
+
+    createSymlink(target, tmp);
+
+    if (rename(tmp.c_str(), link.c_str()) != 0)
+        throw SysError(format("renaming ‘%1%’ to ‘%2%’") % tmp % link);
+}
+
+
 LogType logType = ltPretty;
 Verbosity verbosity = lvlInfo;
 
@@ -1076,6 +1087,15 @@ string chomp(const string & s)
 }
 
 
+string trim(const string & s, const string & whitespace)
+{
+    auto i = s.find_first_not_of(whitespace);
+    if (i == string::npos) return "";
+    auto j = s.find_last_not_of(whitespace);
+    return string(s, i, j == string::npos ? j : j - i + 1);
+}
+
+
 string statusToString(int status)
 {
     if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
diff --git a/src/libutil/util.hh b/src/libutil/util.hh
index 20330fb76..7d20351ee 100644
--- a/src/libutil/util.hh
+++ b/src/libutil/util.hh
@@ -110,6 +110,9 @@ Paths createDirs(const Path & path);
 /* Create a symlink. */
 void createSymlink(const Path & target, const Path & link);
 
+/* Atomically create or replace a symlink. */
+void replaceSymlink(const Path & target, const Path & link);
+
 
 template<class T, class A>
 T singleton(const A & a)
@@ -334,6 +337,10 @@ string concatStringsSep(const string & sep, const StringSet & ss);
 string chomp(const string & s);
 
 
+/* Remove whitespace from the start and end of a string. */
+string trim(const string & s, const string & whitespace = " \n\r\t");
+
+
 /* Convert the exit status of a child as returned by wait() into an
    error string. */
 string statusToString(int status);
diff --git a/src/nix-env/profiles.cc b/src/nix-env/profiles.cc
index d8eb0ef52..d93d7fd80 100644
--- a/src/nix-env/profiles.cc
+++ b/src/nix-env/profiles.cc
@@ -116,16 +116,7 @@ void switchLink(Path link, Path target)
     /* Hacky. */
     if (dirOf(target) == dirOf(link)) target = baseNameOf(target);
 
-    Path tmp = canonPath(dirOf(link) + "/.new_" + baseNameOf(link));
-    createSymlink(target, tmp);
-    /* The rename() system call is supposed to be essentially atomic
-       on Unix.  That is, if we have links `current -> X' and
-       `new_current -> Y', and we rename new_current to current, a
-       process accessing current will see X or Y, but never a
-       file-not-found or other error condition.  This is sufficient to
-       atomically switch user environments. */
-    if (rename(tmp.c_str(), link.c_str()) != 0)
-        throw SysError(format("renaming ‘%1%’ to ‘%2%’") % tmp % link);
+    replaceSymlink(target, link);
 }