chrome_frame/urlmon_bind_status_callback.cc

   1 // Copyright (c) 2011 The Chromium Authors. All rights reserved.
   2 // Use of this source code is governed by a BSD-style license that can be
   3 // found in the LICENSE file.
   4
   5 #include "chrome_frame/urlmon_bind_status_callback.h"
   6
   7 #include <mshtml.h>
   8 #include <shlguid.h>
   9
  10 #include "base/logging.h"
  11 #include "base/string_util.h"
  12 #include "base/stringprintf.h"
  13 #include "base/threading/platform_thread.h"
  14 #include "base/utf_string_conversions.h"
  15 #include "chrome_frame/bind_context_info.h"
  16 #include "chrome_frame/chrome_tab.h"
  17 #include "chrome_frame/exception_barrier.h"
  18 #include "chrome_frame/urlmon_moniker.h"
  19
  20
  21 // A helper to given feed data to the specified |bscb| using
  22 // CacheStream instance.
  23 HRESULT CacheStream::BSCBFeedData(IBindStatusCallback* bscb, const char* data,
  24                                   size_t size, CLIPFORMAT clip_format,
  25                                   size_t flags, bool eof) {
  26   if (!bscb) {
  27     NOTREACHED() << "invalid IBindStatusCallback";
  28     return E_INVALIDARG;
  29   }
  30
  31   // We can't use a CComObjectStackEx here since mshtml will hold
  32   // onto the stream pointer.
  33   CComObject<CacheStream>* cache_stream = NULL;
  34   HRESULT hr = CComObject<CacheStream>::CreateInstance(&cache_stream);
  35   if (FAILED(hr)) {
  36     NOTREACHED();
  37     return hr;
  38   }
  39
  40   scoped_refptr<CacheStream> cache_ref = cache_stream;
  41   hr = cache_stream->Initialize(data, size, eof);
  42   if (FAILED(hr))
  43     return hr;
  44
  45   FORMATETC format_etc = { clip_format, NULL, DVASPECT_CONTENT, -1,
  46                            TYMED_ISTREAM };
  47   STGMEDIUM medium = {0};
  48   medium.tymed = TYMED_ISTREAM;
  49   medium.pstm = cache_stream;
  50
  51   hr = bscb->OnDataAvailable(flags, size, &format_etc, &medium);
  52   return hr;
  53 }
  54
  55 HRESULT CacheStream::Initialize(const char* cache, size_t size, bool eof) {
  56   position_ = 0;
  57   eof_ = eof;
  58
  59   HRESULT hr = S_OK;
  60   cache_.reset(new char[size]);
  61   if (cache_.get()) {
  62     memcpy(cache_.get(), cache, size);
  63     size_ = size;
  64   } else {
  65     DLOG(ERROR) << "failed to allocate cache stream.";
  66     hr = E_OUTOFMEMORY;
  67   }
  68
  69   return hr;
  70 }
  71
  72 // Read is the only call that we expect. Return E_PENDING if there
  73 // is no more data to serve. Otherwise this will result in a
  74 // read with 0 bytes indicating that no more data is available.
  75 STDMETHODIMP CacheStream::Read(void* pv, ULONG cb, ULONG* read) {
  76   if (!pv || !read)
  77     return E_INVALIDARG;
  78
  79   if (!cache_.get()) {
  80     *read = 0;
  81     return S_FALSE;
  82   }
  83
  84   // Default to E_PENDING to signal that this is a partial data.
  85   HRESULT hr = eof_ ? S_FALSE : E_PENDING;
  86   if (position_ < size_) {
  87     *read = std::min(size_ - position_, size_t(cb));
  88     memcpy(pv, cache_ .get() + position_, *read);
  89     position_ += *read;
  90     hr = S_OK;
  91   }
  92
  93   return hr;
  94 }
  95
  96
  97 /////////////////////////////////////////////////////////////////////
  98
  99 HRESULT SniffData::InitializeCache(const std::wstring& url) {
 100   url_ = url;
 101   renderer_type_ = UNDETERMINED;
 102
 103   const int kInitialSize = 4 * 1024; // 4K
 104   HGLOBAL mem = GlobalAlloc(0, kInitialSize);
 105   DCHECK(mem) << "GlobalAlloc failed: " << GetLastError();
 106
 107   HRESULT hr = CreateStreamOnHGlobal(mem, TRUE, cache_.Receive());
 108   if (SUCCEEDED(hr)) {
 109     ULARGE_INTEGER size = {0};
 110     cache_->SetSize(size);
 111   } else {
 112     DLOG(ERROR) << "CreateStreamOnHGlobal failed: " << hr;
 113   }
 114
 115   return hr;
 116 }
 117
 118 HRESULT SniffData::ReadIntoCache(IStream* stream, bool force_determination) {
 119   if (!stream) {
 120     NOTREACHED();
 121     return E_INVALIDARG;
 122   }
 123
 124   HRESULT hr = S_OK;
 125   while (SUCCEEDED(hr)) {
 126     const size_t kChunkSize = 4 * 1024;
 127     char buffer[kChunkSize];
 128     DWORD read = 0;
 129     hr = stream->Read(buffer, sizeof(buffer), &read);
 130     if (read) {
 131       DWORD written = 0;
 132       cache_->Write(buffer, read, &written);
 133       size_ += written;
 134     }
 135
 136     if ((S_FALSE == hr) || !read)
 137       break;
 138   }
 139
 140   bool last_chance = force_determination || (size() >= kMaxSniffSize);
 141   eof_ = force_determination;
 142   DetermineRendererType(last_chance);
 143   return hr;
 144 }
 145
 146 HRESULT SniffData::DrainCache(IBindStatusCallback* bscb, DWORD bscf,
 147                               CLIPFORMAT clip_format) {
 148   if (!is_cache_valid()) {
 149     return S_OK;
 150   }
 151
 152   // Ideally we could just use the cache_ IStream implementation but
 153   // can't use it here since we have to return E_PENDING for the
 154   // last call
 155   HGLOBAL memory = NULL;
 156   HRESULT hr = GetHGlobalFromStream(cache_, &memory);
 157   if (SUCCEEDED(hr) && memory) {
 158     char* buffer = reinterpret_cast<char*>(GlobalLock(memory));
 159     hr = CacheStream::BSCBFeedData(bscb, buffer, size_, clip_format, bscf,
 160                                    eof_);
 161     GlobalUnlock(memory);
 162   }
 163
 164   size_ = 0;
 165   cache_.Release();
 166   return hr;
 167 }
 168
 169 // Scan the buffer or OptIn URL list and decide if the renderer is
 170 // to be switched.  Last chance means there's no more data.
 171 void SniffData::DetermineRendererType(bool last_chance) {
 172   if (is_undetermined()) {
 173     if (last_chance)
 174       renderer_type_ = OTHER;
 175     if (IsChrome(RendererTypeForUrl(url_))) {
 176       renderer_type_ = CHROME;
 177     } else {
 178       if (is_cache_valid() && cache_) {
 179         HGLOBAL memory = NULL;
 180         GetHGlobalFromStream(cache_, &memory);
 181         const char* buffer = reinterpret_cast<const char*>(GlobalLock(memory));
 182
 183         std::wstring html_contents;
 184         // TODO(joshia): detect and handle different content encodings
 185         if (buffer && size_) {
 186           UTF8ToWide(buffer, std::min(size_, kMaxSniffSize), &html_contents);
 187           GlobalUnlock(memory);
 188         }
 189
 190         // Note that document_contents_ may have NULL characters in it. While
 191         // browsers may handle this properly, we don't and will stop scanning
 192         // for the XUACompat content value if we encounter one.
 193         std::wstring xua_compat_content;
 194         UtilGetXUACompatContentValue(html_contents, &xua_compat_content);
 195         if (StrStrI(xua_compat_content.c_str(), kChromeContentPrefix)) {
 196           renderer_type_ = CHROME;
 197         }
 198       }
 199     }
 200     DVLOG(1) << __FUNCTION__ << "Url: " << url_ << base::StringPrintf(
 201           "Renderer type: %s", renderer_type_ == CHROME ? "CHROME" : "OTHER");
 202   }
 203 }
 204
 205 /////////////////////////////////////////////////////////////////////
 206
 207 BSCBStorageBind::BSCBStorageBind() : clip_format_(CF_NULL) {
 208 }
 209
 210 BSCBStorageBind::~BSCBStorageBind() {
 211   std::for_each(saved_progress_.begin(), saved_progress_.end(),
 212                 utils::DeleteObject());
 213 }
 214
 215 HRESULT BSCBStorageBind::Initialize(IMoniker* moniker, IBindCtx* bind_ctx) {
 216   DVLOG(1) << __FUNCTION__ << me()
 217            << base::StringPrintf(" tid=%i", base::PlatformThread::CurrentId());
 218
 219   std::wstring url = GetActualUrlFromMoniker(moniker, bind_ctx,
 220                                              std::wstring());
 221   HRESULT hr = data_sniffer_.InitializeCache(url);
 222   if (FAILED(hr))
 223     return hr;
 224
 225   hr = AttachToBind(bind_ctx);
 226   if (FAILED(hr)) {
 227     NOTREACHED() << __FUNCTION__ << me() << "AttachToBind error: " << hr;
 228     return hr;
 229   }
 230
 231   if (!delegate()) {
 232     NOTREACHED() << __FUNCTION__ << me() << "No existing callback: " << hr;
 233     return E_FAIL;
 234   }
 235
 236   return hr;
 237 }
 238
 239 STDMETHODIMP BSCBStorageBind::OnProgress(ULONG progress, ULONG progress_max,
 240                                     ULONG status_code, LPCWSTR status_text) {
 241   DVLOG(1) << __FUNCTION__ << me()
 242            << base::StringPrintf(" status=%i tid=%i %ls", status_code,
 243                                  base::PlatformThread::CurrentId(),
 244                                  status_text);
 245   // Report all crashes in the exception handler if we wrap the callback.
 246   // Note that this avoids having the VEH report a crash if an SEH earlier in
 247   // the chain handles the exception.
 248   ExceptionBarrier barrier;
 249
 250   HRESULT hr = S_OK;
 251
 252   // TODO(ananta)
 253   // ChromeFrame will not be informed of any redirects which occur while we
 254   // switch into Chrome. This will only break the moniker patch which is
 255   // legacy and needs to be deleted.
 256
 257   if (ShouldCacheProgress(status_code)) {
 258     saved_progress_.push_back(new Progress(progress, progress_max, status_code,
 259                                            status_text));
 260   } else {
 261     hr = CallbackImpl::OnProgress(progress, progress_max, status_code,
 262                                status_text);
 263   }
 264
 265   return hr;
 266 }
 267
 268 // Refer to urlmon_moniker.h for explanation of how things work.
 269 STDMETHODIMP BSCBStorageBind::OnDataAvailable(DWORD flags, DWORD size,
 270                                               FORMATETC* format_etc,
 271                                               STGMEDIUM* stgmed) {
 272   DVLOG(1) << __FUNCTION__
 273            << base::StringPrintf(" tid=%i", base::PlatformThread::CurrentId());
 274   // Report all crashes in the exception handler if we wrap the callback.
 275   // Note that this avoids having the VEH report a crash if an SEH earlier in
 276   // the chain handles the exception.
 277   ExceptionBarrier barrier;
 278   // Do not touch anything other than text/html.
 279   bool is_interesting = (format_etc && stgmed && stgmed->pstm &&
 280       stgmed->tymed == TYMED_ISTREAM &&
 281       IsTextHtmlClipFormat(format_etc->cfFormat));
 282
 283   if (!is_interesting) {
 284     // Play back report progress so far.
 285     MayPlayBack(flags);
 286     return CallbackImpl::OnDataAvailable(flags, size, format_etc, stgmed);
 287   }
 288
 289   HRESULT hr = S_OK;
 290   if (!clip_format_)
 291     clip_format_ = format_etc->cfFormat;
 292
 293   if (data_sniffer_.is_undetermined()) {
 294     bool force_determination = !!(flags &
 295         (BSCF_LASTDATANOTIFICATION | BSCF_DATAFULLYAVAILABLE));
 296     hr = data_sniffer_.ReadIntoCache(stgmed->pstm, force_determination);
 297     // If we don't have sufficient data to determine renderer type
 298     // wait for the next data notification.
 299     if (data_sniffer_.is_undetermined())
 300       return S_OK;
 301   }
 302
 303   DCHECK(!data_sniffer_.is_undetermined());
 304
 305   if (data_sniffer_.is_cache_valid()) {
 306     hr = MayPlayBack(flags);
 307     DCHECK(!data_sniffer_.is_cache_valid());
 308   } else {
 309     hr = CallbackImpl::OnDataAvailable(flags, size, format_etc, stgmed);
 310   }
 311   return hr;
 312 }
 313
 314 STDMETHODIMP BSCBStorageBind::OnStopBinding(HRESULT hresult, LPCWSTR error) {
 315   DVLOG(1) << __FUNCTION__
 316            << base::StringPrintf(" tid=%i", base::PlatformThread::CurrentId());
 317   // Report all crashes in the exception handler if we wrap the callback.
 318   // Note that this avoids having the VEH report a crash if an SEH earlier in
 319   // the chain handles the exception.
 320   ExceptionBarrier barrier;
 321
 322   HRESULT hr = MayPlayBack(BSCF_LASTDATANOTIFICATION);
 323   hr = CallbackImpl::OnStopBinding(hresult, error);
 324   ReleaseBind();
 325   return hr;
 326 }
 327
 328 // Play back the cached data to the delegate. Normally this would happen
 329 // when we have read enough data to determine the renderer. In this case
 330 // we first play back the data from the cache and then go into a 'pass
 331 // through' mode.  In some cases we may end up getting OnStopBinding
 332 // before we get a chance to determine. Also it's possible that the
 333 // BindToStorage call will return before OnStopBinding is sent. Hence
 334 // This is called from 3 places and it's important to maintain the
 335 // exact sequence of calls.
 336 // Once the data is played back, calling this again is a no op.
 337 HRESULT BSCBStorageBind::MayPlayBack(DWORD flags) {
 338   // Force renderer type determination if not already done since
 339   // we want to play back data now.
 340   data_sniffer_.DetermineRendererType(true);
 341   DCHECK(!data_sniffer_.is_undetermined());
 342
 343   HRESULT hr = S_OK;
 344   if (data_sniffer_.is_chrome()) {
 345     // Remember clip format.  If we are switching to chrome, then in order
 346     // to make mshtml return INET_E_TERMINATED_BIND and reissue navigation
 347     // with the same bind context, we have to return a mime type that is
 348     // special cased by mshtml.
 349     static const CLIPFORMAT kMagicClipFormat =
 350         RegisterClipboardFormat(CFSTR_MIME_MPEG);
 351     clip_format_ = kMagicClipFormat;
 352   } else {
 353     if (!saved_progress_.empty()) {
 354       for (ProgressVector::iterator i = saved_progress_.begin();
 355            i != saved_progress_.end(); i++) {
 356         Progress* p = (*i);
 357         // We don't really expect a race condition here but just for sake
 358         // of completeness we check.
 359         if (p) {
 360           (*i) = NULL;
 361           CallbackImpl::OnProgress(p->progress(), p->progress_max(),
 362                                    p->status_code(), p->status_text());
 363           delete p;
 364         }
 365       }
 366       saved_progress_.clear();
 367     }
 368   }
 369
 370   if (data_sniffer_.is_cache_valid()) {
 371     if (data_sniffer_.is_chrome()) {
 372       base::win::ScopedComPtr<BindContextInfo> info;
 373       BindContextInfo::FromBindContext(bind_ctx_, info.Receive());
 374       DCHECK(info);
 375       if (info) {
 376         info->SetToSwitch(data_sniffer_.cache_);
 377       }
 378     }
 379
 380     hr = data_sniffer_.DrainCache(delegate(),
 381         flags | BSCF_FIRSTDATANOTIFICATION, clip_format_);
 382     DLOG_IF(WARNING, INET_E_TERMINATED_BIND != hr) << __FUNCTION__ <<
 383       " mshtml OnDataAvailable returned: " << std::hex << hr;
 384   }
 385
 386   return hr;
 387 }
 388
 389 // We cache and suppress sending progress notifications till
 390 // we get the first OnDataAvailable. This is to prevent
 391 // mshtml from making up its mind about the mime type.
 392 // However, this is the invasive part of the patch and
 393 // could trip other software that's due to mistimed progress
 394 // notifications. It is probably not a good idea to hide redirects
 395 // and some cookie notifications.
 396 //
 397 // We only need to suppress data notifications like
 398 // BINDSTATUS_MIMETYPEAVAILABLE,
 399 // BINDSTATUS_CACHEFILENAMEAVAILABLE etc.
 400 //
 401 // This is an atempt to reduce the exposure by starting to
 402 // cache only when we receive one of the interesting progress
 403 // notification.
 404 bool BSCBStorageBind::ShouldCacheProgress(unsigned long status_code) const {
 405   // We need to cache progress notifications only if we haven't yet figured
 406   // out which way the request is going.
 407   if (data_sniffer_.is_undetermined()) {
 408     // If we are already caching then continue.
 409     if (!saved_progress_.empty())
 410       return true;
 411     // Start caching only if we see one of the interesting progress
 412     // notifications.
 413     switch (status_code) {
 414       case BINDSTATUS_BEGINDOWNLOADDATA:
 415       case BINDSTATUS_DOWNLOADINGDATA:
 416       case BINDSTATUS_USINGCACHEDCOPY:
 417       case BINDSTATUS_MIMETYPEAVAILABLE:
 418       case BINDSTATUS_CACHEFILENAMEAVAILABLE:
 419       case BINDSTATUS_SERVER_MIMETYPEAVAILABLE:
 420         return true;
 421       default:
 422         break;
 423     }
 424   }
 425
 426   return false;
 427 }