| #ifndef THIRD_PARTY_GBMCWEB_INCLUDE_WEBASSETS_H_ | |
| #define THIRD_PARTY_GBMCWEB_INCLUDE_WEBASSETS_H_ | |
| #include <algorithm> | |
| #include <array> | |
| #include <cstring> | |
| #include <filesystem> // NOLINT | |
| #include <fstream> | |
| #include <iterator> | |
| #include <memory> | |
| #include <string> | |
| #include <system_error> // NOLINT | |
| #include <utility> | |
| #include <vector> | |
| #include "boost/algorithm/string/replace.hpp" // NOLINT | |
| #include "boost/container/flat_set.hpp" // NOLINT | |
| #include "app.hpp" | |
| #include "http_request.hpp" | |
| #include "http_response.hpp" | |
| #include "logging.hpp" | |
| #include "routing.hpp" | |
| #include "async_resp.hpp" | |
| #include "forward_unauthorized.hpp" | |
| #include "webroutes.hpp" | |
| namespace crow { | |
| namespace webassets { | |
| struct CmpStr { | |
| bool operator()(const char* a, const char* b) const { | |
| return std::strcmp(a, b) < 0; | |
| } | |
| }; | |
| inline void requestRoutes(App& app) { | |
| constexpr static std::array<std::pair<const char*, const char*>, 17> | |
| contentTypes{ | |
| {{".css", "text/css;charset=UTF-8"}, | |
| {".html", "text/html;charset=UTF-8"}, | |
| {".js", "application/javascript;charset=UTF-8"}, | |
| {".png", "image/png;charset=UTF-8"}, | |
| {".woff", "application/x-font-woff"}, | |
| {".woff2", "application/x-font-woff2"}, | |
| {".gif", "image/gif"}, | |
| {".ico", "image/x-icon"}, | |
| {".ttf", "application/x-font-ttf"}, | |
| {".svg", "image/svg+xml"}, | |
| {".eot", "application/vnd.ms-fontobject"}, | |
| {".xml", "application/xml"}, | |
| {".json", "application/json"}, | |
| {".jpg", "image/jpeg"}, | |
| {".jpeg", "image/jpeg"}, | |
| // dev tools don't care about map type, setting to json causes | |
| // browser to show as text | |
| // https://stackoverflow.com/questions/19911929/what-mime-type-should-i-use-for-javascript-source-map-files | |
| {".map", "application/json"}}}; | |
| std::filesystem::path rootpath{"/usr/share/www/"}; | |
| std::error_code ec; | |
| std::filesystem::recursive_directory_iterator dirIter(rootpath, ec); | |
| if (ec) { | |
| BMCWEB_LOG_ERROR << "Unable to find or open " << rootpath.string() | |
| << " static file hosting disabled"; | |
| return; | |
| } | |
| // In certain cases, we might have both a gzipped version of the file AND a | |
| // non-gzipped version. To avoid duplicated routes, we need to make sure we | |
| // get the gzipped version first. Because the gzipped path should be longer | |
| // than the non gzipped path, if we sort in descending order, we should be | |
| // guaranteed to get the gzip version first. | |
| std::vector<std::filesystem::directory_entry> paths( | |
| std::filesystem::begin(dirIter), std::filesystem::end(dirIter)); | |
| std::sort(paths.rbegin(), paths.rend()); | |
| for (const std::filesystem::directory_entry& dir : paths) { | |
| const std::filesystem::path& absolutePath = dir.path(); | |
| std::filesystem::path relativePath{ | |
| absolutePath.string().substr(rootpath.string().size() - 1)}; | |
| if (std::filesystem::is_directory(dir)) { | |
| // don't recurse into hidden directories or symlinks | |
| if (dir.path().filename().string().starts_with(".") || | |
| std::filesystem::is_symlink(dir)) { | |
| dirIter.disable_recursion_pending(); | |
| } | |
| } else if (std::filesystem::is_regular_file(dir)) { | |
| std::string extension = relativePath.extension(); | |
| std::filesystem::path webpath = relativePath; | |
| const char* contentEncoding = nullptr; | |
| if (extension == ".gz") { | |
| webpath = webpath.replace_extension(""); | |
| // Use the non-gzip version for determining content type | |
| extension = webpath.extension().string(); | |
| contentEncoding = "gzip"; | |
| } | |
| if (webpath.filename().string().starts_with("index.")) { | |
| webpath = webpath.parent_path(); | |
| if (webpath.string().empty() || webpath.string().back() != '/') { | |
| // insert the non-directory version of this path | |
| webroutes::routes.insert(webpath); | |
| webpath += "/"; | |
| } | |
| } | |
| std::pair<boost::container::flat_set<std::string>::iterator, bool> | |
| inserted = webroutes::routes.insert(webpath); | |
| if (!inserted.second) { | |
| // Got a duplicated path. This is expected in certain | |
| // situations | |
| BMCWEB_LOG_DEBUG << "Got duplicated path " << webpath.string(); | |
| continue; | |
| } | |
| const char* contentType = nullptr; | |
| for (const std::pair<const char*, const char*>& ext : contentTypes) { | |
| if (ext.first == nullptr || ext.second == nullptr) { | |
| continue; | |
| } | |
| if (extension == ext.first) { | |
| contentType = ext.second; | |
| } | |
| } | |
| if (contentType == nullptr) { | |
| BMCWEB_LOG_ERROR << "Cannot determine content-type for " | |
| << absolutePath.string() << " with extension " | |
| << extension; | |
| } | |
| if (webpath == "/") { | |
| forward_unauthorized::hasWebuiRoute = true; | |
| } | |
| app.routeDynamic(webpath)( | |
| [absolutePath, contentType, contentEncoding]( | |
| const crow::Request&, | |
| const std::shared_ptr<bmcweb::AsyncResp>& asyncResp) { | |
| if (contentType != nullptr) { | |
| asyncResp->res.addHeader(boost::beast::http::field::content_type, | |
| contentType); | |
| } | |
| if (contentEncoding != nullptr) { | |
| asyncResp->res.addHeader( | |
| boost::beast::http::field::content_encoding, contentEncoding); | |
| } | |
| // res.set_header("Cache-Control", "public, max-age=86400"); | |
| std::ifstream inf(absolutePath); | |
| if (!inf) { | |
| BMCWEB_LOG_DEBUG << "failed to read file"; | |
| asyncResp->res.result( | |
| boost::beast::http::status::internal_server_error); | |
| return; | |
| } | |
| asyncResp->res.body() = {std::istreambuf_iterator<char>(inf), | |
| std::istreambuf_iterator<char>()}; | |
| }); | |
| } | |
| } | |
| } | |
| } // namespace webassets | |
| } // namespace crow | |
| #endif // THIRD_PARTY_GBMCWEB_INCLUDE_WEBASSETS_H_ |