Files
lix/lix/libutil/archive.cc
T
eldritch horrorsandRaito Bezarius 2f7d7444f1 libutil: fix nar parser buffer overflow
string data shares a buffer with the binary string length field. size
calculations for string read buffers always include the length field;
sufficiently large length fields can cause these calculations to wrap.
a malicious nar could use this for OOB writes in the daemon (as root).

since we use strings only as tags for archive members and for symlinks
with their OS-dependent length limits we can simply limit string size.
1 MiB should be sufficient for all symlinks, and tags are always tiny.

Change-Id: I89fb05f73c1dbeda45d91244aba4cd526a3d83e1
2026-05-04 19:00:19 +02:00

1104 lines
37 KiB
C++

#include <cerrno>
#include <algorithm>
#include <string_view>
#include <vector>
#include <map>
#include <strings.h> // for strcasecmp
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>
#include <dirent.h>
#include <fcntl.h>
#include "lix/libutil/archive.hh"
#include "c-calls.hh"
#include "lix/libutil/async-io.hh"
#include "lix/libutil/box_ptr.hh"
#include "lix/libutil/config.hh"
#include "lix/libutil/file-system.hh"
#include "lix/libutil/finally.hh"
#include "lix/libutil/generator.hh"
#include "lix/libutil/logging.hh"
#include "lix/libutil/result.hh"
#include "lix/libutil/serialise.hh"
#include "lix/libutil/signals.hh"
namespace nix {
struct ArchiveSettings : Config
{
#include "archive-settings.gen.inc"
};
static ArchiveSettings realArchiveSettings;
Config & archiveSettings = realArchiveSettings;
PathFilter defaultPathFilter = [](const Path &) { return true; };
static WireFormatGenerator dumpContents(Path path, off_t size)
{
AutoCloseFD fd{sys::open(path, O_RDONLY | O_CLOEXEC)};
if (!fd) throw SysError("opening file '%1%'", path);
std::vector<char> buf(65536);
size_t left = size;
while (left > 0) {
auto n = std::min(left, buf.size());
readFull(fd.get(), buf.data(), n);
left -= n;
co_yield std::span{buf.data(), n};
}
}
static WireFormatGenerator dumpSingle(nar::File f)
{
co_yield "type";
co_yield "regular";
if (f.executable) {
co_yield "executable";
co_yield "";
}
co_yield "contents";
co_yield f.size;
co_yield std::move(f.contents);
co_yield SerializingTransform::padding(f.size);
}
static WireFormatGenerator dumpSingle(nar::Symlink s)
{
co_yield "type";
co_yield "symlink";
co_yield "target";
co_yield s.target;
}
static WireFormatGenerator dumpSingle(nar::Directory d)
{
co_yield "type";
co_yield "directory";
while (auto e = d.contents.next()) {
co_yield std::visit(
[&](auto & i) {
return [](auto & name, auto & i) -> WireFormatGenerator {
co_yield "entry";
co_yield "(";
co_yield "name";
co_yield name;
co_yield "node";
co_yield "(";
co_yield dumpSingle(std::move(i));
co_yield ")";
co_yield ")";
}(e->first, i);
},
e->second
);
}
}
WireFormatGenerator nar::dump(nar::Entry nar)
{
co_yield narVersionMagic1;
co_yield "(";
co_yield std::visit(
[](auto i) -> WireFormatGenerator { return dumpSingle(std::move(i)); }, std::move(nar)
);
co_yield ")";
}
// list the given path under the given filter and return the oldest mtime.
// if returnUnhacked is true directory entries that appear to have had the
// nix case hack applied will be returned without the case hack suffix, if
// returnUnhacked is false directory entries will be returned as they have
// been read from disk. to produce a correct NAR from the results the case
// hack must be undone if configured unless returnUnhacked is set to true.
static nar::Entry list(Path path, time_t & mtime, PathFilter & filter, bool returnUnhacked)
{
checkInterrupt();
auto st = lstat(path);
mtime = st.st_mtime;
if (S_ISREG(st.st_mode)) {
return nar::File{
(st.st_mode & S_IXUSR) != 0,
uint64_t(st.st_size),
dumpContents(std::move(path), st.st_size)
};
} else if (S_ISDIR(st.st_mode)) {
auto contents = [](Path path, time_t & mtime, PathFilter & filter, bool returnUnhacked
) -> Generator<std::pair<const std::string &, nar::Entry>> {
/* If we're on a case-insensitive system like macOS, undo
the case hack applied by restorePath(). */
std::map<std::string, std::string> unhacked;
for (auto & i : readDirectory(path))
// See Note [Case Hack].
if (realArchiveSettings.useCaseHack) {
std::string name(i.name);
size_t pos = i.name.find(caseHackSuffix);
if (pos != std::string::npos) {
debug("removing case hack suffix from '%1%'", path + "/" + i.name);
name.erase(pos);
}
if (!unhacked.emplace(name, i.name).second)
throw Error("file name collision in between '%1%' and '%2%'",
(path + "/" + unhacked[name]),
(path + "/" + i.name));
} else {
unhacked.emplace(i.name, i.name);
}
for (auto & i : unhacked) {
if (filter(path + "/" + i.first)) {
time_t tmp_mtime;
auto diskPath = path + "/" + i.second;
co_yield std::pair(
std::cref(returnUnhacked ? i.first : i.second),
list(diskPath, tmp_mtime, filter, returnUnhacked)
);
if (tmp_mtime > mtime) {
mtime = tmp_mtime;
}
}
}
};
return nar::Directory(contents(std::move(path), mtime, filter, returnUnhacked));
} else if (S_ISLNK(st.st_mode)) {
return nar::Symlink{readLink(path)};
} else {
throw Error("file '%1%' has an unsupported type", path);
}
}
WireFormatGenerator dumpPathAndGetMtime(Path path, time_t & mtime)
{
co_yield dump(list(path, mtime, defaultPathFilter, true));
}
WireFormatGenerator dumpPath(Path path, PathFilter & filter)
{
auto filtered = prepareDump(std::move(path), filter);
co_yield filtered->dump();
}
WireFormatGenerator dumpPath(Path path)
{
auto prepared = prepareDump(std::move(path));
co_yield prepared->dump();
}
WireFormatGenerator dumpString(std::string_view s)
{
co_yield narVersionMagic1;
co_yield "(";
co_yield "type";
co_yield "regular";
co_yield "contents";
co_yield s;
co_yield ")";
}
struct UnfilteredDump : PreparedDump
{
using PreparedDump::PreparedDump;
WireFormatGenerator dump() const override
{
time_t ignored;
co_yield nar::dump(list(rootPath, ignored, defaultPathFilter, true));
}
};
struct PrefilteredDump : PreparedDump
{
struct File
{
bool executable;
uint64_t size;
};
struct Symlink
{
Path target;
};
struct Directory;
using Entry = std::variant<File, Symlink, Directory>;
struct Directory
{
std::vector<std::pair<std::string, Entry>> contents;
};
Entry root;
PrefilteredDump(Path path, PathFilter & filter) : PreparedDump(std::move(path))
{
time_t ignored;
fillFrom(root, list(rootPath, ignored, filter, false));
}
static void fillFrom(Entry & target, nar::Entry e)
{
overloaded handlers{
[&](nar::File & f) { target = File{f.executable, f.size}; },
[&](nar::Symlink & s) { target = Symlink{std::move(s.target)}; },
[&](nar::Directory & d) {
Directory self;
while (auto entry = d.contents.next()) {
fillFrom(
self.contents.emplace_back(std::move(entry->first), Entry{}).second,
std::move(entry->second)
);
}
target = std::move(self);
},
};
std::visit(handlers, e);
}
static nar::Entry convert(Path path, const Entry & e)
{
overloaded handlers{
[&](const File & f) -> nar::Entry {
return nar::File{f.executable, f.size, dumpContents(std::move(path), f.size)};
},
[&](const Symlink & s) -> nar::Entry { return nar::Symlink{s.target}; },
[&](const Directory & d) -> nar::Entry {
return nar::Directory{
[](Path path, const Directory & d
) -> Generator<std::pair<const std::string &, nar::Entry>> {
for (auto & [name, entry] : d.contents) {
// FIXME(jade): what?! we have two copies of this case un-hack code?
std::string narName = realArchiveSettings.useCaseHack
? name.substr(0, name.find(caseHackSuffix))
: name;
co_yield std::pair{
std::cref(narName), convert(path + "/" + name, entry)
};
}
}(std::move(path), d)
};
},
};
return std::visit(handlers, std::move(e));
}
WireFormatGenerator dump() const override
{
return nar::dump(convert(rootPath, root));
}
};
box_ptr<PreparedDump> prepareDump(Path path)
{
return make_box_ptr<UnfilteredDump>(std::move(path));
}
box_ptr<PreparedDump> prepareDump(Path path, PathFilter & filter)
{
return make_box_ptr<PrefilteredDump>(std::move(path), filter);
}
static SerialisationError badArchive(const std::string & s)
{
return SerialisationError("bad archive: " + s);
}
struct CaseInsensitiveCompare
{
bool operator() (const std::string & a, const std::string & b) const
{
// NOLINTNEXTLINE(lix-unsafe-c-calls): valid pathnames never contain nuls
return strcasecmp(a.c_str(), b.c_str()) < 0;
}
};
namespace nar {
namespace {
struct Parser
{
struct FileHeader
{
bool executable;
uint64_t size;
};
struct Symlink
{
Path target;
};
struct Directory;
struct WantBytes
{
size_t n;
};
using Response = std::variant<FileHeader, Symlink, Directory, WantBytes>;
struct Directory
{
using Entry = std::pair<const Path &, Generator<Response>>;
using Stream = Generator<std::variant<WantBytes, Entry>>;
Stream content;
};
std::vector<char> & buffer;
// these macros purposely duplicate parts of the wire protocol,
// but in such a way that doing it *wrong* will definitely make
// tests fail. we could also duplicate them completely, but not
// doing so ensures that we're the inverse of dump at all times
#define FETCH_INT(type) \
({ \
co_yield WantBytes{8}; \
StringSource src(std::string_view(buffer.data(), buffer.size())); \
readNum<type>(src); \
})
#define READ_U64() \
({ \
auto u = FETCH_INT(uint64_t); \
buffer.clear(); \
u; \
})
#define READ_STRING_LIMITED(limit) \
({ \
size_t len = FETCH_INT(size_t); \
if (len > (limit)) { \
throw SerialisationError( \
"found malformed string tag. input may be a compressed NAR, which cannot be read " \
"directly" \
); \
} \
co_yield WantBytes{len + (8 - len % 8) % 8}; \
StringSource src(std::string_view(buffer.data(), buffer.size())); \
auto str = readString(src, (limit)); \
buffer.clear(); \
std::move(str); \
})
#define READ_STRING() READ_STRING_LIMITED(1048576)
#define READ_PADDING(size) \
do { \
if ((size) % 8) { \
co_yield WantBytes{size_t(8 - (size) % 8)}; \
StringSource src(std::string_view(buffer.data(), buffer.size())); \
readPadding((size), src); \
buffer.clear(); \
} \
} while (0)
#define EXPECT(raw, kind) \
do { \
auto s = READ_STRING(); \
if (s != (raw)) { \
throw badArchive("expected " kind " tag"); \
} \
} while (0)
Generator<Response> parse()
{
EXPECT("(", "open");
EXPECT("type", "type");
const auto t = READ_STRING();
if (t == "regular") {
auto contentsOrFlag = READ_STRING();
const bool executable = contentsOrFlag == "executable";
if (executable) {
auto s = READ_STRING();
if (s != "") {
throw badArchive("executable marker has non-empty value");
}
contentsOrFlag = READ_STRING();
}
if (contentsOrFlag == "contents") {
const uint64_t size = READ_U64();
co_yield FileHeader{executable, size};
READ_PADDING(size);
} else {
throw badArchive("file without contents found");
}
} else if (t == "directory") {
// NOLINTNEXTLINE(cppcoreguidelines-avoid-capturing-lambda-coroutines)
auto makeReader = [this](bool & completed) -> Directory::Stream {
std::string prevName;
while (1) {
{
const auto s = READ_STRING();
if (s == ")") {
break;
} else if (s != "entry") {
throw badArchive("expected entry tag");
}
EXPECT("(", "open");
}
EXPECT("name", "name");
auto name = READ_STRING();
if (name.empty() || name == "." || name == ".."
|| name.find('/') != std::string::npos
|| name.find((char) 0) != std::string::npos
// The case hack is a thing that only exists on the
// filesystem.
// Unpacking one appearing in a NAR is super
// sketchy because it will at minimum cause corruption at
// the time of repacking the NAR.
|| name.find(caseHackSuffix) != std::string::npos)
{
throw Error("NAR contains invalid file name '%1%'", name);
}
if (name <= prevName) {
throw Error("NAR directory is not sorted");
}
prevName = name;
// N.B. The restore visitor will case-hack the filename if necessary
// See Note [Case Hack].
EXPECT("node", "node");
co_yield Directory::Entry{name, parse()};
EXPECT(")", "close");
}
completed = true;
};
bool completed = false;
co_yield Directory{makeReader(completed)};
// directories may nest, so to drain a directory properly we'd have to add a Finally
// argument to the generator to ensure that the draining code is always run. this is
// usually not necessary, hard to follow, and rather error-prone on top of all that.
assert(completed);
// directories are terminated already, don't try to read another ")"
co_return;
} else if (t == "symlink") {
EXPECT("target", "target");
std::string target = READ_STRING();
co_yield Symlink{target};
} else {
throw badArchive("unknown file type " + t);
}
EXPECT(")", "close");
}
Generator<Response> parseRoot()
{
std::string version;
try {
version = READ_STRING_LIMITED(narVersionMagic1.size());
if (version != narVersionMagic1) {
throw SerialisationError("bad NAR version tag");
}
co_yield parse();
} catch (SerialisationError & e) {
throw badArchive(fmt("input doesn't look like a Nix archive (%s)", e.info().msg.str()));
}
}
#undef FETCH_INT
#undef READ_U64
#undef READ_STRING
#undef READ_STRING_LIMITED
#undef READ_PADDING
#undef EXPECT
};
struct SyncParser
{
Source & source;
std::vector<char> buffer;
Generator<Entry> parse()
{
Parser parser{buffer};
auto stream = parser.parseRoot();
co_yield parse(stream);
}
void feed(size_t n)
{
checkInterrupt();
auto end = buffer.size();
buffer.resize(end + n);
source(buffer.data() + end, n);
}
Generator<std::pair<const std::string &, Entry>>
readDir(Parser::Directory::Stream stream)
{
while (auto e = stream.next()) {
if (auto want = std::get_if<Parser::WantBytes>(&*e)) {
feed(want->n);
} else if (auto entry = std::get_if<Parser::Directory::Entry>(&*e)) {
auto parsed = parse(entry->second);
while (auto e = parsed.next()) {
co_yield std::pair(std::cref(entry->first), std::move(*e));
}
} else {
assert(false && "expected parser response in dir");
}
}
}
Generator<Entry> parse(Generator<Parser::Response> & stream)
{
while (auto i = stream.next()) {
if (auto want = std::get_if<Parser::WantBytes>(&*i)) {
feed(want->n);
} else if (auto f = std::get_if<Parser::FileHeader>(&*i)) {
auto makeReader = [](Source & source, uint64_t & left) -> Generator<Bytes> {
std::array<char, 65536> buf;
while (left) {
checkInterrupt();
auto n = size_t(std::min<uint64_t>(buf.size(), left));
source(buf.data(), n);
co_yield std::span{buf.data(), n};
left -= n;
}
};
auto left = f->size;
co_yield File{f->executable, f->size, makeReader(source, left)};
// we could drain the remainder of the file, but coroutines being interruptible
// at any time makes this difficult. for files this is not that hard, but being
// consistent with directories is more important than handling the simple case.
assert(left == 0);
} else if (auto sl = std::get_if<Parser::Symlink>(&*i)) {
co_yield Symlink{std::move(sl->target)};
} else if (auto dir = std::get_if<Parser::Directory>(&*i)) {
co_yield Directory{readDir(std::move(dir->content))};
} else {
assert(false && "unhandled parser response");
}
}
}
};
struct AsyncCopier : AsyncInputStream
{
AsyncInputStream & source;
std::vector<char> buffer;
Parser parser{buffer};
struct Fragment
{
// how many bytes the parser requested but we haven't read from source yet
uint64_t pending = 0;
// whether the requested bytes are nar metadata (false) or contents (true)
bool pendingFileContents = false;
};
Generator<Fragment> stream{ignoreContents(parser.parseRoot())};
Fragment current;
explicit AsyncCopier(AsyncInputStream & source) : source(source) {}
kj::Promise<Result<std::optional<size_t>>> read(void * buffer, size_t size) override
try {
while (current.pending == 0) {
if (auto want = stream.next()) {
current = *want;
} else {
co_return std::nullopt;
}
}
size = std::min<uint64_t>(current.pending, size);
if (size == 0) {
co_return std::nullopt;
}
auto got = TRY_AWAIT(source.read(buffer, size));
if (!got) {
throw badArchive("truncated NAR encountered");
}
current.pending -= *got;
if (!current.pendingFileContents) {
auto end = this->buffer.size();
this->buffer.resize(end + *got);
memcpy(this->buffer.data() + end, buffer, *got);
}
co_return *got;
} catch (...) {
co_return result::current_exception();
}
Generator<Fragment> ignoreContents(Generator<Parser::Response> stream)
{
while (auto i = stream.next()) {
if (auto want = std::get_if<Parser::WantBytes>(&*i)) {
co_yield Fragment{want->n, false};
} else if (auto f = std::get_if<Parser::FileHeader>(&*i)) {
co_yield Fragment{f->size, true};
} else if (auto _ = std::get_if<Parser::Symlink>(&*i)) {
// nothing to do
} else if (auto dir = std::get_if<Parser::Directory>(&*i)) {
while (auto e = dir->content.next()) {
if (auto want = std::get_if<Parser::WantBytes>(&*e)) {
co_yield Fragment{want->n, false};
} else if (auto entry = std::get_if<Parser::Directory::Entry>(&*e)) {
co_yield ignoreContents(std::move(entry->second));
} else {
assert(false && "expected parser response in dir");
}
}
} else {
assert(false && "unhandled parser response");
}
}
}
};
// sadly async parsers can't be written to produce a tree of generators
// the way sync parsers can. once we have async generators that may not
// be as hard, but async generators in kj might have too much overhead.
struct AsyncParser
{
AsyncInputStream & source;
std::vector<char> buffer;
kj::Promise<Result<void>> parse(NARParseVisitor & target)
try {
Parser parser{buffer};
auto stream = parser.parseRoot();
TRY_AWAIT(parse(stream, target, ""));
co_return result::success();
} catch (...) {
co_return result::current_exception();
}
kj::Promise<Result<void>> read(char * buffer, size_t n)
try {
if (!TRY_AWAIT(source.readRange(buffer, n, n))) {
throw badArchive("unexpected end of nar encountered");
}
co_return result::success();
} catch (...) {
co_return result::current_exception();
}
kj::Promise<Result<void>> feed(size_t n)
{
auto end = buffer.size();
buffer.resize(end + n);
return read(buffer.data() + end, n);
}
kj::Promise<Result<void>>
parse(Generator<Parser::Response> & stream, NARParseVisitor & target, const std::string & name)
try {
while (auto i = stream.next()) {
if (auto want = std::get_if<Parser::WantBytes>(&*i)) {
TRY_AWAIT(feed(want->n));
} else if (auto f = std::get_if<Parser::FileHeader>(&*i)) {
auto file = target.createRegularFile(name, f->size, f->executable);
auto left = f->size;
std::array<char, 65536> buf;
while (left) {
auto n = size_t(std::min<uint64_t>(buf.size(), left));
TRY_AWAIT(read(buf.data(), n));
file->receiveContents({buf.data(), n});
left -= n;
}
file->close();
} else if (auto sl = std::get_if<Parser::Symlink>(&*i)) {
target.createSymlink(name, sl->target);
} else if (auto d = std::get_if<Parser::Directory>(&*i)) {
auto dir = target.createDirectory(name);
while (auto e = d->content.next()) {
if (auto want = std::get_if<Parser::WantBytes>(&*e)) {
TRY_AWAIT(feed(want->n));
} else if (auto entry = std::get_if<Parser::Directory::Entry>(&*e)) {
TRY_AWAIT(parse(entry->second, *dir, entry->first));
} else {
assert(false && "expected parser response in dir");
}
}
} else {
assert(false && "unhandled parser response");
}
}
co_return result::success();
} catch (...) {
co_return result::current_exception();
}
};
}
Generator<Entry> parse(Source & source)
{
SyncParser p{source};
co_yield p.parse();
}
}
namespace nar_index {
namespace {
struct NarPositioner
{
uint64_t pos = 0;
};
struct Indexer : NARParseVisitor
{
NarPositioner & source;
Directory & parent;
public:
Indexer(NarPositioner & source, Directory & parent) : source(source), parent(parent) {}
box_ptr<NARParseVisitor> createDirectory(const std::string & name) override
{
auto & dir = std::get<Directory>(parent.contents[name] = Directory{});
return make_box_ptr<Indexer>(source, dir);
}
box_ptr<FileHandle>
createRegularFile(const std::string & name, uint64_t size, bool executable) override
{
struct IgnoringFileHandle : FileHandle
{
void close() override {}
void receiveContents(std::string_view data) override {}
};
parent.contents[name] = File{executable, source.pos, size};
return make_box_ptr<IgnoringFileHandle>();
}
void createSymlink(const std::string & name, const std::string & target) override
{
parent.contents[name] = Symlink{target};
}
};
}
Entry create(Source & source)
{
struct NarSource : Source, NarPositioner
{
Source & source;
NarSource(Source & source) : source(source) {}
size_t read(char * data, size_t len) override
{
auto n = source.read(data, len);
pos += n;
return n;
}
};
Directory root;
NarSource wrapper{source};
Indexer index{wrapper, root};
parseDump(index, wrapper);
return root.contents.at("");
}
kj::Promise<Result<Entry>> create(AsyncInputStream & source)
try {
struct NarSource : AsyncInputStream, NarPositioner
{
AsyncInputStream & source;
NarSource(AsyncInputStream & source) : source(source) {}
kj::Promise<Result<std::optional<size_t>>> read(void * data, size_t len) override
try {
auto n = TRY_AWAIT(source.read(data, len));
pos += n.value_or(0);
co_return n;
} catch (...) {
co_return result::current_exception();
}
};
Directory root;
NarSource wrapper{source};
Indexer index{wrapper, root};
TRY_AWAIT(parseDump(index, wrapper));
co_return root.contents.at("");
} catch (...) {
co_return result::current_exception();
}
}
static void restore(NARParseVisitor & sink, nar::Entry entry, const Path & path)
{
return std::visit(
overloaded{
[&](nar::File f) {
auto handle = sink.createRegularFile(path, f.size, f.executable);
while (auto block = f.contents.next()) {
handle->receiveContents(std::string_view{block->data(), block->size()});
}
handle->close();
},
[&](nar::Symlink sl) { sink.createSymlink(path, sl.target); },
[&](nar::Directory d) {
auto dir = sink.createDirectory(path);
while (auto entry = d.contents.next()) {
restore(*dir, std::move(entry->second), entry->first);
}
},
},
std::move(entry)
);
}
void parseDump(NARParseVisitor & sink, Source & source)
{
auto nar = nar::parse(source);
while (auto entry = nar.next()) {
restore(sink, std::move(*entry), "");
}
}
kj::Promise<Result<void>> parseDump(NARParseVisitor & sink, AsyncInputStream & source)
try {
nar::AsyncParser parser{source};
TRY_AWAIT(parser.parse(sink));
co_return result::success();
} catch (...) {
co_return result::current_exception();
}
/* Note [Case Hack]:
* Nix uses a "case hack" which intentionally messes up filenames of files that
* have conflicts only in case so that the mapping to a case insensitive
* filesystem is one-to-one and data is not corrupted when loading the data
* from said filesystem to send to a case-sensitive one.
*
* It exists so that NARs with case conflicts can be successfully extracted on
* default macOS installations and then re-compressed and sent to Linux
* machines without corrupting them.
*
* For example, a NAR with the files "pod" and "Pod" will extract as:
* - Pod
* - pod~nix~case~hack~1
*
* The case hacked filenames consist of a magic string `caseHackSuffix`, which
* is `~nix~case~hack~`, then an increasing number based on the number of
* conflicts that file name has.
*
* However: this is ITSELF corruption of NARs and is the cause of numerous
* bugs, and to top it off, it is not necessary anymore in a world where the
* Nix store is already on a separate APFS container *anyway*, so we can just
* enable case sensitivity on macOS and remove the case hack.
*
* It is *already* the case that not all NARs that exist can be extracted on
* macOS without throwing an extraction error; see
* Note [NAR restoration security]: Unicode normalization conflicts already
* error today.
*
* Unlike HFS+, APFS never corrupts filenames: it does not unicode-normalize
* them and will give you out the same output as you put in; HOWEVER, it uses a
* Unicode normalization insensitive hash function when searching for them,
* which means that two files with the same name under Unicode NFD will resolve
* to the same underlying file and fail as per Note [NAR restoration security].
*
* Lix intends to remove the case hack, see:
* https://git.lix.systems/lix-project/lix/issues/332
* https://git.lix.systems/lix-project/lix/projects/16
*/
/*
* Note [NAR restoration security]:
* It's *critical* that NAR restoration will never overwrite anything even if
* duplicate filenames are passed in. It is inevitable that not all NARs are
* fit to actually successfully restore to the target filesystem; errors may
* occur due to collisions, and this *must* cause the NAR to be rejected.
*
* Although the filenames are blocked from being *the same bytes* by a higher
* layer, filesystems have other ideas on every platform:
* - The store may be on a case-insensitive filesystem like APFS, ext4 with
* casefold directories, zfs with casesensitivity=insensitive
* - The store may be on a Unicode normalizing (or normalization-insensitive)
* filesystem like APFS (where files are looked up by
* hash(normalize(fname))), HFS+ (where file names are always normalized to
* approximately NFD), or zfs with normalization=formC, etc.
*
* It is impossible to know the version of Unicode being used by the underlying
* filesystem, thus it is *impossible* to stop these collisions.
*
* Overwriting files as a result of invalid NARs will cause a security bug like
* CppNix's CVE-2024-45593 (GHSA-h4vv-h3jq-v493)
*/
/**
* This code restores NARs from disk.
*
* See Note [NAR restoration security] for security invariants in this procedure.
*
*/
struct NARRestoreVisitor : NARParseVisitor
{
Path dstPath;
bool useCaseHack;
std::map<Path, int, CaseInsensitiveCompare> caseHackNames;
private:
struct MyFileHandle : public FileHandle
{
AutoCloseFD fd;
MyFileHandle(AutoCloseFD && fd, uint64_t size, bool executable) : FileHandle(), fd(std::move(fd))
{
if (executable) {
makeExecutable();
}
maybePreallocateContents(size);
}
void makeExecutable()
{
struct stat st;
if (fstat(fd.get(), &st) == -1)
throw SysError("fstat");
if (fchmod(fd.get(), st.st_mode | (S_IXUSR | S_IXGRP | S_IXOTH)) == -1)
throw SysError("fchmod");
}
void maybePreallocateContents(uint64_t len)
{
if (!realArchiveSettings.preallocateContents) {
return;
}
#if HAVE_POSIX_FALLOCATE
if (len) {
errno = posix_fallocate(fd.get(), 0, len);
/* Note that EINVAL may indicate that the underlying
filesystem doesn't support preallocation (e.g. on
OpenSolaris). Since preallocation is just an
optimisation, ignore it. */
if (errno && errno != EINVAL && errno != EOPNOTSUPP && errno != ENOSYS)
throw SysError("preallocating file of %1% bytes", len);
}
#endif
}
~MyFileHandle() = default;
virtual void close() override
{
/* Call close explicitly to make sure the error is checked */
fd.close();
}
void receiveContents(std::string_view data) override
{
writeFull(fd.get(), data);
}
};
/** See Note [Case Hack] */
std::string maybeCaseHackFilename(std::string const & name)
{
if (this->useCaseHack) {
auto i = caseHackNames.find(name);
if (i != caseHackNames.end()) {
debug("case collision between '%1%' and '%2%'", i->first, name);
auto name2 = name;
name2 += caseHackSuffix;
name2 += std::to_string(++i->second);
return name2;
} else {
caseHackNames[name] = 0;
}
}
return name;
}
public:
NARRestoreVisitor(Path dstPath, bool useCaseHack): dstPath(std::move(dstPath)), useCaseHack(useCaseHack) {}
box_ptr<NARParseVisitor> createDirectory(const std::string & name_) override
{
auto name = maybeCaseHackFilename(name_);
Path p = dstPath + name;
if (sys::mkdir(p, 0777) == -1) {
throw SysError("creating directory '%1%'", p);
}
return make_box_ptr<NARRestoreVisitor>(p + "/", useCaseHack);
};
box_ptr<FileHandle> createRegularFile(const std::string & name_, uint64_t size, bool executable) override
{
auto name = maybeCaseHackFilename(name_);
Path p = dstPath + name;
AutoCloseFD fd = sys::open(p, O_CREAT | O_EXCL | O_WRONLY | O_CLOEXEC, 0666);
if (!fd) throw SysError("creating file '%1%'", p);
return make_box_ptr<MyFileHandle>(std::move(fd), size, executable);
}
void createSymlink(const std::string & name_, const std::string & target) override
{
auto name = maybeCaseHackFilename(name_);
Path p = dstPath + name;
nix::createSymlink(target, p);
}
};
void restorePath(const Path & path, Source & source)
{
NARRestoreVisitor sink(path, realArchiveSettings.useCaseHack);
parseDump(sink, source);
}
kj::Promise<Result<void>> restorePath(const Path & path, AsyncInputStream & source)
try {
NARRestoreVisitor sink(path, realArchiveSettings.useCaseHack);
TRY_AWAIT(parseDump(sink, source));
co_return result::success();
} catch (...) {
co_return result::current_exception();
}
WireFormatGenerator copyNAR(Source & source)
{
// FIXME: if 'source' is the output of dumpPath() followed by EOF,
// we should just forward all data directly without parsing.
auto items = nar::parse(source);
// we can't use dump() here because we must read the entire nar *before*
// returning the final `)` tag, otherwise the source will not be emptied
// before the returned generator is exhausted. that in turn confuses the
// remote store protocols that expect copyNAR to not finish any earlier.
co_yield narVersionMagic1;
co_yield "(";
for (auto && item : items) {
co_yield std::visit([](auto i) { return dumpSingle(std::move(i)); }, std::move(item));
}
co_yield ")";
}
box_ptr<AsyncInputStream> copyNAR(AsyncInputStream & source)
{
return make_box_ptr<nar::AsyncCopier>(source);
}
}