Source code

Revision control

Copy as Markdown

Other Tools

/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
/* vim: set ts=8 sts=2 et sw=2 tw=80: */
/* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
#include "URL.h"
#include "URLMainThread.h"
#include "URLWorker.h"
#include "nsASCIIMask.h"
#include "MainThreadUtils.h"
#include "mozilla/RefPtr.h"
#include "mozilla/dom/URLBinding.h"
#include "mozilla/dom/BindingUtils.h"
#include "nsContentUtils.h"
#include "mozilla/dom/Document.h"
#include "nsIURIMutator.h"
#include "nsNetUtil.h"
namespace mozilla::dom {
NS_IMPL_CYCLE_COLLECTION_WRAPPERCACHE(URL, mParent, mSearchParams)
NS_IMPL_CYCLE_COLLECTING_ADDREF(URL)
NS_IMPL_CYCLE_COLLECTING_RELEASE(URL)
NS_INTERFACE_MAP_BEGIN_CYCLE_COLLECTION(URL)
NS_WRAPPERCACHE_INTERFACE_MAP_ENTRY
NS_INTERFACE_MAP_ENTRY(nsISupports)
NS_INTERFACE_MAP_END
JSObject* URL::WrapObject(JSContext* aCx, JS::Handle<JSObject*> aGivenProto) {
return URL_Binding::Wrap(aCx, this, aGivenProto);
}
/* static */
already_AddRefed<URL> URL::Constructor(const GlobalObject& aGlobal,
const nsACString& aURL,
const Optional<nsACString>& aBase,
ErrorResult& aRv) {
if (aBase.WasPassed()) {
return Constructor(aGlobal.GetAsSupports(), aURL, aBase.Value(), aRv);
}
return Constructor(aGlobal.GetAsSupports(), aURL, nullptr, aRv);
}
/* static */
already_AddRefed<URL> URL::Constructor(nsISupports* aParent,
const nsACString& aURL,
const nsACString& aBase,
ErrorResult& aRv) {
nsCOMPtr<nsIURI> baseUri;
nsresult rv = NS_NewURI(getter_AddRefs(baseUri), aBase);
if (NS_WARN_IF(NS_FAILED(rv))) {
aRv.ThrowTypeError<MSG_INVALID_URL>(aBase);
return nullptr;
}
return Constructor(aParent, aURL, baseUri, aRv);
}
/* static */
already_AddRefed<URL> URL::Constructor(nsISupports* aParent,
const nsACString& aURL, nsIURI* aBase,
ErrorResult& aRv) {
nsCOMPtr<nsIURI> uri;
nsresult rv = NS_NewURI(getter_AddRefs(uri), aURL, nullptr, aBase);
if (NS_FAILED(rv)) {
// No need to warn in this case. It's common to use the URL constructor
// to determine if a URL is valid and an exception will be propagated.
aRv.ThrowTypeError<MSG_INVALID_URL>(aURL);
return nullptr;
}
return MakeAndAddRef<URL>(aParent, std::move(uri));
}
already_AddRefed<URL> URL::FromURI(GlobalObject& aGlobal, nsIURI* aURI) {
return MakeAndAddRef<URL>(aGlobal.GetAsSupports(), aURI);
}
void URL::CreateObjectURL(const GlobalObject& aGlobal, Blob& aBlob,
nsACString& aResult, ErrorResult& aRv) {
if (NS_IsMainThread()) {
URLMainThread::CreateObjectURL(aGlobal, aBlob, aResult, aRv);
} else {
URLWorker::CreateObjectURL(aGlobal, aBlob, aResult, aRv);
}
}
void URL::CreateObjectURL(const GlobalObject& aGlobal, MediaSource& aSource,
nsACString& aResult, ErrorResult& aRv) {
MOZ_ASSERT(NS_IsMainThread());
URLMainThread::CreateObjectURL(aGlobal, aSource, aResult, aRv);
}
void URL::RevokeObjectURL(const GlobalObject& aGlobal, const nsACString& aURL,
ErrorResult& aRv) {
if (aURL.Contains('#')) {
// Don't revoke URLs that contain fragments.
return;
}
if (NS_IsMainThread()) {
URLMainThread::RevokeObjectURL(aGlobal, aURL, aRv);
} else {
URLWorker::RevokeObjectURL(aGlobal, aURL, aRv);
}
}
bool URL::IsValidObjectURL(const GlobalObject& aGlobal, const nsACString& aURL,
ErrorResult& aRv) {
if (NS_IsMainThread()) {
return URLMainThread::IsValidObjectURL(aGlobal, aURL, aRv);
}
return URLWorker::IsValidObjectURL(aGlobal, aURL, aRv);
}
already_AddRefed<nsIURI> URL::ParseURI(const nsACString& aURL,
const Optional<nsACString>& aBase) {
nsCOMPtr<nsIURI> baseUri;
nsCOMPtr<nsIURI> uri;
if (aBase.WasPassed()) {
nsresult rv = NS_NewURI(getter_AddRefs(baseUri), aBase.Value());
if (NS_FAILED(rv)) {
return nullptr;
}
}
nsresult rv = NS_NewURI(getter_AddRefs(uri), aURL, nullptr, baseUri);
if (NS_FAILED(rv)) {
return nullptr;
}
return uri.forget();
};
already_AddRefed<URL> URL::Parse(const GlobalObject& aGlobal,
const nsACString& aURL,
const Optional<nsACString>& aBase) {
nsCOMPtr<nsIURI> uri = ParseURI(aURL, aBase);
if (!uri) {
return nullptr;
}
return MakeAndAddRef<URL>(aGlobal.GetAsSupports(), std::move(uri));
}
bool URL::CanParse(const GlobalObject& aGlobal, const nsACString& aURL,
const Optional<nsACString>& aBase) {
nsCOMPtr<nsIURI> uri = ParseURI(aURL, aBase);
return !!uri;
}
URLSearchParams* URL::SearchParams() {
CreateSearchParamsIfNeeded();
return mSearchParams;
}
bool IsChromeURI(nsIURI* aURI) { return aURI->SchemeIs("chrome"); }
void URL::CreateSearchParamsIfNeeded() {
if (!mSearchParams) {
mSearchParams = new URLSearchParams(mParent, this);
UpdateURLSearchParams();
}
}
void URL::SetSearch(const nsACString& aSearch) {
SetSearchInternal(aSearch);
UpdateURLSearchParams();
}
void URL::URLSearchParamsUpdated(URLSearchParams* aSearchParams) {
MOZ_ASSERT(mSearchParams);
MOZ_ASSERT(mSearchParams == aSearchParams);
nsAutoCString search;
mSearchParams->Serialize(search);
SetSearchInternal(search);
}
#define URL_GETTER(value, func) \
MOZ_ASSERT(mURI); \
mURI->func(value);
void URL::GetHref(nsACString& aHref) const { URL_GETTER(aHref, GetSpec); }
void URL::SetHref(const nsACString& aHref, ErrorResult& aRv) {
nsCOMPtr<nsIURI> uri;
nsresult rv = NS_NewURI(getter_AddRefs(uri), aHref);
if (NS_FAILED(rv)) {
aRv.ThrowTypeError<MSG_INVALID_URL>(aHref);
return;
}
mURI = std::move(uri);
UpdateURLSearchParams();
}
void URL::GetOrigin(nsACString& aOrigin) const {
nsresult rv =
nsContentUtils::GetWebExposedOriginSerialization(URI(), aOrigin);
if (NS_WARN_IF(NS_FAILED(rv))) {
aOrigin.Truncate();
}
}
void URL::GetProtocol(nsACString& aProtocol) const {
URL_GETTER(aProtocol, GetScheme);
aProtocol.Append(char16_t(':'));
}
void URL::SetProtocol(const nsACString& aProtocol) {
nsCOMPtr<nsIURI> uri(URI());
if (!uri) {
return;
}
uri = net::TryChangeProtocol(uri, aProtocol);
if (!uri) {
return;
}
mURI = std::move(uri);
}
void URL::GetUsername(nsACString& aUsername) const {
URL_GETTER(aUsername, GetUsername);
}
void URL::SetUsername(const nsACString& aUsername) {
MOZ_ASSERT(mURI);
Unused << NS_MutateURI(mURI).SetUsername(aUsername).Finalize(mURI);
}
void URL::GetPassword(nsACString& aPassword) const {
URL_GETTER(aPassword, GetPassword);
}
void URL::SetPassword(const nsACString& aPassword) {
MOZ_ASSERT(mURI);
Unused << NS_MutateURI(mURI).SetPassword(aPassword).Finalize(mURI);
}
void URL::GetHost(nsACString& aHost) const { URL_GETTER(aHost, GetHostPort); }
void URL::SetHost(const nsACString& aHost) {
MOZ_ASSERT(mURI);
Unused << NS_MutateURI(mURI).SetHostPort(aHost).Finalize(mURI);
}
void URL::GetHostname(nsACString& aHostname) const {
MOZ_ASSERT(mURI);
aHostname.Truncate();
nsContentUtils::GetHostOrIPv6WithBrackets(mURI, aHostname);
}
void URL::SetHostname(const nsACString& aHostname) {
MOZ_ASSERT(mURI);
// nsStandardURL returns NS_ERROR_UNEXPECTED for an empty hostname
// The return code is silently ignored
Unused << NS_MutateURI(mURI).SetHost(aHostname).Finalize(mURI);
}
void URL::GetPort(nsACString& aPort) const {
MOZ_ASSERT(mURI);
aPort.Truncate();
int32_t port;
nsresult rv = mURI->GetPort(&port);
if (NS_SUCCEEDED(rv) && port != -1) {
aPort.AppendInt(port, 10);
}
}
void URL::SetPort(const nsACString& aPort) {
nsresult rv;
nsAutoCString portStr(aPort);
int32_t port = -1;
// nsIURI uses -1 as default value.
portStr.StripTaggedASCII(ASCIIMask::MaskCRLFTab());
if (!portStr.IsEmpty()) {
// To be valid, the port must start with an ASCII digit.
// (nsACString::ToInteger ignores leading junk, so check before calling.)
if (!IsAsciiDigit(portStr[0])) {
return;
}
port = portStr.ToInteger(&rv);
if (NS_FAILED(rv)) {
return;
}
}
Unused << NS_MutateURI(mURI).SetPort(port).Finalize(mURI);
}
void URL::GetPathname(nsACString& aPathname) const {
MOZ_ASSERT(mURI);
// Do not throw! Not having a valid URI or URL should result in an empty
// string.
mURI->GetFilePath(aPathname);
}
void URL::SetPathname(const nsACString& aPathname) {
MOZ_ASSERT(mURI);
// Do not throw!
Unused << NS_MutateURI(mURI).SetFilePath(aPathname).Finalize(mURI);
}
void URL::GetSearch(nsACString& aSearch) const {
MOZ_ASSERT(mURI);
aSearch.Truncate();
// Do not throw! Not having a valid URI or URL should result in an empty
// string.
nsresult rv;
rv = mURI->GetQuery(aSearch);
if (NS_SUCCEEDED(rv) && !aSearch.IsEmpty()) {
aSearch.Insert('?', 0);
}
}
void URL::GetHash(nsACString& aHash) const {
MOZ_ASSERT(mURI);
aHash.Truncate();
nsresult rv = mURI->GetRef(aHash);
if (NS_SUCCEEDED(rv) && !aHash.IsEmpty()) {
aHash.Insert('#', 0);
}
}
void URL::SetHash(const nsACString& aHash) {
MOZ_ASSERT(mURI);
Unused << NS_MutateURI(mURI).SetRef(aHash).Finalize(mURI);
}
void URL::SetSearchInternal(const nsACString& aSearch) {
MOZ_ASSERT(mURI);
// Ignore failures to be compatible with NS4.
Unused << NS_MutateURI(mURI).SetQuery(aSearch).Finalize(mURI);
}
void URL::UpdateURLSearchParams() {
if (!mSearchParams) {
return;
}
nsAutoCString search;
nsresult rv = URI()->GetQuery(search);
if (NS_WARN_IF(NS_FAILED(rv))) {
search.Truncate();
}
mSearchParams->ParseInput(search);
}
nsIURI* URL::URI() const {
MOZ_ASSERT(mURI);
return mURI;
}
} // namespace mozilla::dom