Merge pull request #14597 from NixOS/restore-sink-openat

libutil: Make RestoreSink use *at system calls on UNIX

src/libutil/archive.cc

@@ -200,54 +200,54 @@ static void parse(FileSystemObjectSink & sink, Source & source, const CanonPath
     }
 
     else if (type == "directory") {
-        sink.createDirectory(path);
+        sink.createDirectory(path, [&](FileSystemObjectSink & dirSink, const CanonPath & relDirPath) {
+            std::map<Path, int, CaseInsensitiveCompare> names;
 
-        std::map<Path, int, CaseInsensitiveCompare> names;
+            std::string prevName;
 
-        std::string prevName;
+            while (1) {
+                auto tag = getString();
 
-        while (1) {
-            auto tag = getString();
+                if (tag == ")")
+                    break;
 
-            if (tag == ")")
-                break;
+                if (tag != "entry")
+                    throw badArchive("expected tag 'entry' or ')', got '%s'", tag);
 
-            if (tag != "entry")
-                throw badArchive("expected tag 'entry' or ')', got '%s'", tag);
+                expectTag("(");
 
-            expectTag("(");
+                expectTag("name");
 
-            expectTag("name");
+                auto name = getString();
+                if (name.empty() || name == "." || name == ".." || name.find('/') != std::string::npos
+                    || name.find((char) 0) != std::string::npos)
+                    throw badArchive("NAR contains invalid file name '%1%'", name);
+                if (name <= prevName)
+                    throw badArchive("NAR directory is not sorted");
+                prevName = name;
+                if (archiveSettings.useCaseHack) {
+                    auto i = names.find(name);
+                    if (i != names.end()) {
+                        debug("case collision between '%1%' and '%2%'", i->first, name);
+                        name += caseHackSuffix;
+                        name += std::to_string(++i->second);
+                        auto j = names.find(name);
+                        if (j != names.end())
+                            throw badArchive(
+                                "NAR contains file name '%s' that collides with case-hacked file name '%s'",
+                                prevName,
+                                j->first);
+                    } else
+                        names[name] = 0;
+                }
 
-            auto name = getString();
-            if (name.empty() || name == "." || name == ".." || name.find('/') != std::string::npos
-                || name.find((char) 0) != std::string::npos)
-                throw badArchive("NAR contains invalid file name '%1%'", name);
-            if (name <= prevName)
-                throw badArchive("NAR directory is not sorted");
-            prevName = name;
-            if (archiveSettings.useCaseHack) {
-                auto i = names.find(name);
-                if (i != names.end()) {
-                    debug("case collision between '%1%' and '%2%'", i->first, name);
-                    name += caseHackSuffix;
-                    name += std::to_string(++i->second);
-                    auto j = names.find(name);
-                    if (j != names.end())
-                        throw badArchive(
-                            "NAR contains file name '%s' that collides with case-hacked file name '%s'",
-                            prevName,
-                            j->first);
-                } else
-                    names[name] = 0;
+                expectTag("node");
+
+                parse(dirSink, source, relDirPath / name);
+
+                expectTag(")");
             }
-
-            expectTag("node");
-
-            parse(sink, source, path / name);
-
-            expectTag(")");
-        }
+        });
     }
 
     else if (type == "symlink") {

src/libutil/fs-sink.cc

@@ -34,11 +34,12 @@ void copyRecursive(SourceAccessor & accessor, const CanonPath & from, FileSystem
     }
 
     case SourceAccessor::tDirectory: {
-        sink.createDirectory(to);
-        for (auto & [name, _] : accessor.readDirectory(from)) {
-            copyRecursive(accessor, from / name, sink, to / name);
-            break;
-        }
+        sink.createDirectory(to, [&](FileSystemObjectSink & dirSink, const CanonPath & relDirPath) {
+            for (auto & [name, _] : accessor.readDirectory(from)) {
+                copyRecursive(accessor, from / name, dirSink, relDirPath / name);
+                break;
+            }
+        });
         break;
     }
 
@@ -70,11 +71,60 @@ static std::filesystem::path append(const std::filesystem::path & src, const Can
     return dst;
 }
 
+#ifndef _WIN32
+void RestoreSink::createDirectory(const CanonPath & path, DirectoryCreatedCallback callback)
+{
+    if (path.isRoot()) {
+        createDirectory(path);
+        callback(*this, path);
+        return;
+    }
+
+    createDirectory(path);
+    assert(dirFd); // If that's not true the above call must have thrown an exception.
+
+    RestoreSink dirSink{startFsync};
+    dirSink.dstPath = append(dstPath, path);
+    dirSink.dirFd = ::openat(dirFd.get(), path.rel_c_str(), O_RDONLY | O_DIRECTORY | O_NOFOLLOW | O_CLOEXEC);
+
+    if (!dirSink.dirFd)
+        throw SysError("opening directory '%s'", dirSink.dstPath.string());
+
+    callback(dirSink, CanonPath::root);
+}
+#endif
+
 void RestoreSink::createDirectory(const CanonPath & path)
 {
     auto p = append(dstPath, path);
+
+#ifndef _WIN32
+    if (dirFd) {
+        if (path.isRoot())
+            /* Trying to create a directory that we already have a file descriptor for. */
+            throw Error("path '%s' already exists", p.string());
+
+        if (::mkdirat(dirFd.get(), path.rel_c_str(), 0777) == -1)
+            throw SysError("creating directory '%s'", p.string());
+
+        return;
+    }
+#endif
+
     if (!std::filesystem::create_directory(p))
         throw Error("path '%s' already exists", p.string());
+
+#ifndef _WIN32
+    if (path.isRoot()) {
+        assert(!dirFd); // Handled above
+
+        /* Open directory for further *at operations relative to the sink root
+           directory. */
+        dirFd = open(p.c_str(), O_RDONLY | O_DIRECTORY | O_NOFOLLOW | O_CLOEXEC);
+        if (!dirFd)
+            throw SysError("creating directory '%1%'", p.string());
+    }
+#endif
 };
 
 struct RestoreRegularFile : CreateRegularFileSink
@@ -114,7 +164,14 @@ void RestoreSink::createRegularFile(const CanonPath & path, std::function<void(C
             FILE_ATTRIBUTE_NORMAL,
             NULL)
 #else
-        open(p.c_str(), O_CREAT | O_EXCL | O_WRONLY | O_CLOEXEC, 0666)
+        [&]() {
+            /* O_EXCL together with O_CREAT ensures symbolic links in the last
+              component are not followed. */
+            constexpr int flags = O_CREAT | O_EXCL | O_WRONLY | O_CLOEXEC;
+            if (!dirFd)
+                return ::open(p.c_str(), flags, 0666);
+            return ::openat(dirFd.get(), path.rel_c_str(), flags, 0666);
+        }();
 #endif
         ;
     if (!crf.fd)
@@ -161,6 +218,13 @@ void RestoreRegularFile::operator()(std::string_view data)
 void RestoreSink::createSymlink(const CanonPath & path, const std::string & target)
 {
     auto p = append(dstPath, path);
+#ifndef _WIN32
+    if (dirFd) {
+        if (::symlinkat(requireCString(target), dirFd.get(), path.rel_c_str()) == -1)
+            throw SysError("creating symlink from '%1%' -> '%2%'", p.string(), target);
+        return;
+    }
+#endif
     nix::createSymlink(target, p.string());
 }
 

src/libutil/include/nix/util/fs-sink.hh

@@ -36,6 +36,23 @@ struct FileSystemObjectSink
 
     virtual void createDirectory(const CanonPath & path) = 0;
 
+    using DirectoryCreatedCallback = std::function<void(FileSystemObjectSink & dirSink, const CanonPath & dirRelPath)>;
+
+    /**
+     * Create a directory and invoke a callback with a pair of sink + CanonPath
+     * of the created subdirectory relative to dirSink.
+     *
+     * @note This allows for UNIX RestoreSink implementations to implement
+     * *at-style accessors that always keep an open file descriptor for the
+     * freshly created directory. Use this when it's important to disallow any
+     * intermediate path components from being symlinks.
+     */
+    virtual void createDirectory(const CanonPath & path, DirectoryCreatedCallback callback)
+    {
+        createDirectory(path);
+        callback(*this, path);
+    }
+
     /**
      * This function in general is no re-entrant. Only one file can be
      * written at a time.
@@ -82,6 +99,18 @@ struct NullFileSystemObjectSink : FileSystemObjectSink
 struct RestoreSink : FileSystemObjectSink
 {
     std::filesystem::path dstPath;
+#ifndef _WIN32
+    /**
+     * File descriptor for the directory located at dstPath. Used for *at
+     * operations relative to this file descriptor. This sink must *never*
+     * follow intermediate symlinks (starting from dstPath) in case a file
+     * collision is encountered for various reasons like case-insensitivity or
+     * other types on normalization. using appropriate *at system calls and traversing
+     * only one path component at a time ensures that writing is race-free and is
+     * is not susceptible to symlink replacement.
+     */
+    AutoCloseFD dirFd;
+#endif
     bool startFsync = false;
 
     explicit RestoreSink(bool startFsync)
@@ -91,6 +120,10 @@ struct RestoreSink : FileSystemObjectSink
 
     void createDirectory(const CanonPath & path) override;
 
+#ifndef _WIN32
+    void createDirectory(const CanonPath & path, DirectoryCreatedCallback callback) override;
+#endif
+
     void createRegularFile(const CanonPath & path, std::function<void(CreateRegularFileSink &)>) override;
 
     void createSymlink(const CanonPath & path, const std::string & target) override;

src/libutil/include/nix/util/strings.hh

@@ -166,4 +166,10 @@ public:
     }
 };
 
+/**
+ * Check that the string does not contain any NUL bytes and return c_str().
+ * @throws Error if str contains '\0' bytes.
+ */
+const char * requireCString(const std::string & str);
+
 } // namespace nix

src/libutil/strings.cc

@@ -5,6 +5,7 @@
 #include "nix/util/strings-inline.hh"
 #include "nix/util/os-string.hh"
 #include "nix/util/error.hh"
+#include "nix/util/util.hh"
 
 namespace nix {
 
@@ -152,4 +153,14 @@ std::string optionalBracket(std::string_view prefix, std::string_view content, s
     return result;
 }
 
+const char * requireCString(const std::string & s)
+{
+    if (std::memchr(s.data(), '\0', s.size())) [[unlikely]] {
+        using namespace std::string_view_literals;
+        auto str = replaceStrings(s, "\0"sv, "␀"sv);
+        throw Error("string '%s' with null (\\0) bytes used where it's not allowed", str);
+    }
+    return s.c_str();
+}
+
 } // namespace nix

src/libutil/url.cc

@@ -327,8 +327,11 @@ Path renderUrlPathEnsureLegal(const std::vector<std::string> & urlPath)
         /* This is only really valid for UNIX. Windows has more restrictions. */
         if (comp.contains('/'))
             throw BadURL("URL path component '%s' contains '/', which is not allowed in file names", comp);
-        if (comp.contains(char(0)))
-            throw BadURL("URL path component '%s' contains NUL byte which is not allowed", comp);
+        if (comp.contains(char(0))) {
+            using namespace std::string_view_literals;
+            auto str = replaceStrings(comp, "\0"sv, "␀"sv);
+            throw BadURL("URL path component '%s' contains NUL byte which is not allowed", str);
+        }
     }
 
     return concatStringsSep("/", urlPath);

tests/functional/nars.sh

@@ -114,7 +114,7 @@ if (( unicodeTestCode == 1 )); then
     # If the command failed (MacOS or ZFS + normalization), checks that it failed
     # with the expected "already exists" error, and that this is the same
     # behavior as `touch`
-    echo "$unicodeTestOut" | grepQuiet "path '.*/out/â' already exists"
+    echo "$unicodeTestOut" | grepQuiet "creating directory '.*/out/â': File exists"
 
     (( touchFilesCount == 1 ))
 elif (( unicodeTestCode == 0 )); then