Support hooking so in an archive

lsplt/src/main/jni/include/lsplt.hpp

@@ -41,28 +41,63 @@ struct MapInfo {
     [[maybe_unused, gnu::visibility("default")]] static std::vector<MapInfo> Scan();
 };
 
-/// \brief Register a hook to a function.
+/// \brief Register a hook to a function by inode. For so within an archive, you should use
+/// #RegisterHook(ino_t, uintptr_t, size_t, std::string_view, void *, void **) instead.
 /// \param inode The inode of the library to hook. You can obtain the inode by #stat() or by finding
-/// the library in the list returned by #MapInfo::Scan().
+/// the library in the list returned by #lsplt::v1::MapInfo::Scan().
 /// \param symbol The function symbol to hook.
 /// \param callback The callback function pointer to call when the function is called.
-/// \param backup The backup function pointer which can call the original function. This is optional.
+/// \param backup The backup function pointer which can call the original function. This is
+/// optional.
 /// \return Whether the hook is successfully registered.
 /// \note This function is thread-safe.
 /// \note \p backup will not be available until #CommitHook() is called.
 /// \note \p backup will be nullptr if the hook fails.
 /// \note You can unhook the function by calling this function with \p callback set to the backup
 /// set by previous call.
-/// \note LSPlt will backup the hook memory region and restore it when the hook is restored
+/// \note LSPlt will backup the hook memory region and restore it when the
-/// to its original function pointer so that there won't be dirty pages.
+/// hook is restored to its original function pointer so that there won't be dirty pages. LSPlt will
-/// LSPlt will do hooks on a copied memory region so that the original memory region will not be
+/// do hooks on a copied memory region so that the original memory region will not be modified. You
-/// modified. You can invalidate this behaviour and hook the original memory region by calling
+/// can invalidate this behaviour and hook the original memory region by calling
 /// #InvalidateBackup().
 /// \see #CommitHook()
 /// \see #InvalidateBackup()
 [[maybe_unused, gnu::visibility("default")]] bool RegisterHook(ino_t inode, std::string_view symbol,
                                                                void *callback, void **backup);
 
+/// \brief Register a hook to a function by inode with offset range. This is useful when hooking
+/// a library that is directly loaded from an archive without extraction.
+/// \param inode The inode of the library to hook. You can obtain the inode by #stat() or by finding
+/// the library in the list returned by #lsplt::v1::MapInfo::Scan().
+/// \param offset The to the library in the file.
+/// \param size The upper bound size to the library in the file.
+/// \param symbol The function symbol to hook.
+/// \param callback The callback function pointer to call when the function is called.
+/// \param backup The backup function pointer which can call the original function. This is
+/// optional.
+/// \return Whether the hook is successfully registered.
+/// \note This function is thread-safe.
+/// \note \p backup will not be available until #CommitHook() is called.
+/// \note \p backup will be nullptr if the hook fails.
+/// \note You can unhook the function by calling this function with \p callback set to the backup
+/// set by previous call.
+/// \note LSPlt will backup the hook memory region and restore it when the
+/// hook is restored to its original function pointer so that there won't be dirty pages. LSPlt will
+/// do hooks on a copied memory region so that the original memory region will not be modified. You
+/// can invalidate this behaviour and hook the original memory region by calling
+/// #InvalidateBackup().
+/// \note You can get the offset range of the library by getting its entry offset and size in the
+/// zip file.
+/// \note According to the Android linker specification, the \p offset must be page aligned.
+/// \note The \p offset must be accurate, otherwise the hook may fail because the ELF header
+/// cannot be found.
+/// \note The \p size can be inaccurate but should be larger or equal to the library size,
+/// otherwise the hook may fail when the hook pointer is beyond the range.
+/// \see #CommitHook()
+/// \see #InvalidateBackup()
+[[maybe_unused, gnu::visibility("default")]] bool RegisterHook(ino_t inode, uintptr_t offset,
+                                                               size_t size, std::string_view symbol,
+                                                               void *callback, void **backup);
 /// \brief Commit all registered hooks.
 /// \return Whether all hooks are successfully committed. If any of the hooks fail to commit,
 /// the result is false.

lsplt/src/main/jni/lsplt.cc

@@ -24,6 +24,7 @@ inline auto PageEnd(uintptr_t addr) {
 
 struct RegisterInfo {
     ino_t inode;
+    std::pair<uintptr_t, uintptr_t> offset_range;
     std::string symbol;
     void *callback;
     void **backup;
@@ -34,7 +35,10 @@ struct HookInfo : public lsplt::MapInfo {
     uintptr_t backup;
     std::unique_ptr<Elf> elf;
     bool self;
-    [[nodiscard]] bool Match(const RegisterInfo &info) const { return info.inode == inode; }
+    [[nodiscard]] bool Match(const RegisterInfo &info) const {
+        return info.inode == inode && offset >= info.offset_range.first &&
+               offset < info.offset_range.second;
+    }
 };
 
 class HookInfos : public std::map<uintptr_t, HookInfo, std::greater<>> {
@@ -172,7 +176,7 @@ public:
         for (auto &[_, info] : *this) {
             for (auto iter = register_info.begin(); iter != register_info.end();) {
                 const auto &reg = *iter;
-                if (info.offset != 0 || !info.Match(reg)) {
+                if (info.offset != iter->offset_range.first || !info.Match(reg)) {
                     ++iter;
                     continue;
                 }
@@ -268,11 +272,33 @@ inline namespace v1 {
 [[maybe_unused]] bool RegisterHook(ino_t inode, std::string_view symbol, void *callback,
                                    void **backup) {
     if (inode == 0 || symbol.empty() || !callback) return false;
-    LOGV("RegisterHook %lu %s", inode, symbol.data());
 
     std::unique_lock lock(hook_mutex);
-    register_info.emplace_back(RegisterInfo{inode, std::string{symbol}, callback, backup});
+    static_assert(std::numeric_limits<uintptr_t>::min() == 0);
+    static_assert(std::numeric_limits<uintptr_t>::max() == -1);
+    const auto &info = register_info.emplace_back(
+        RegisterInfo{inode,
+                     {std::numeric_limits<uintptr_t>::min(), std::numeric_limits<uintptr_t>::max()},
+                     std::string{symbol},
+                     callback,
+                     backup});
 
+    LOGV("RegisterHook %lu %s", info.inode, info.symbol.data());
+    return true;
+}
+
+[[maybe_unused]] bool RegisterHook(ino_t inode, uintptr_t offset, size_t size,
+                                   std::string_view symbol, void *callback, void **backup) {
+    if (inode == 0 || symbol.empty() || !callback) return false;
+
+    std::unique_lock lock(hook_mutex);
+    static_assert(std::numeric_limits<uintptr_t>::min() == 0);
+    static_assert(std::numeric_limits<uintptr_t>::max() == -1);
+    const auto &info = register_info.emplace_back(
+        RegisterInfo{inode, {offset, offset + size}, std::string{symbol}, callback, backup});
+
+    LOGV("RegisterHook %lu %" PRIxPTR "-%" PRIxPTR " %s", info.inode, info.offset_range.first,
+         info.offset_range.second, info.symbol.data());
     return true;
 }