net/proxy/proxy_script_fetcher_impl.cc

   1 // Copyright (c) 2012 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 "net/proxy/proxy_script_fetcher_impl.h"
   6
   7 #include "base/compiler_specific.h"
   8 #include "base/location.h"
   9 #include "base/logging.h"
  10 #include "base/metrics/histogram.h"
  11 #include "base/single_thread_task_runner.h"
  12 #include "base/strings/string_util.h"
  13 #include "base/thread_task_runner_handle.h"
  14 #include "net/base/data_url.h"
  15 #include "net/base/io_buffer.h"
  16 #include "net/base/load_flags.h"
  17 #include "net/base/net_errors.h"
  18 #include "net/base/net_string_util.h"
  19 #include "net/base/request_priority.h"
  20 #include "net/cert/cert_status_flags.h"
  21 #include "net/http/http_response_headers.h"
  22 #include "net/url_request/url_request_context.h"
  23
  24 // TODO(eroman):
  25 //   - Support auth-prompts (http://crbug.com/77366)
  26
  27 namespace net {
  28
  29 namespace {
  30
  31 // The maximum size (in bytes) allowed for a PAC script. Responses exceeding
  32 // this will fail with ERR_FILE_TOO_BIG.
  33 const int kDefaultMaxResponseBytes = 1048576;  // 1 megabyte
  34
  35 // The maximum duration (in milliseconds) allowed for fetching the PAC script.
  36 // Responses exceeding this will fail with ERR_TIMED_OUT.
  37 const int kDefaultMaxDurationMs = 300000;  // 5 minutes
  38
  39 // Returns true if |mime_type| is one of the known PAC mime type.
  40 bool IsPacMimeType(const std::string& mime_type) {
  41   static const char * const kSupportedPacMimeTypes[] = {
  42     "application/x-ns-proxy-autoconfig",
  43     "application/x-javascript-config",
  44   };
  45   for (size_t i = 0; i < arraysize(kSupportedPacMimeTypes); ++i) {
  46     if (base::LowerCaseEqualsASCII(mime_type, kSupportedPacMimeTypes[i]))
  47       return true;
  48   }
  49   return false;
  50 }
  51
  52 // Converts |bytes| (which is encoded by |charset|) to UTF16, saving the resul
  53 // to |*utf16|.
  54 // If |charset| is empty, then we don't know what it was and guess.
  55 void ConvertResponseToUTF16(const std::string& charset,
  56                             const std::string& bytes,
  57                             base::string16* utf16) {
  58   const char* codepage;
  59
  60   if (charset.empty()) {
  61     // Assume ISO-8859-1 if no charset was specified.
  62     codepage = kCharsetLatin1;
  63   } else {
  64     // Otherwise trust the charset that was provided.
  65     codepage = charset.c_str();
  66   }
  67
  68   // Be generous in the conversion -- if any characters lie outside of |charset|
  69   // (i.e. invalid), then substitute them with U+FFFD rather than failing.
  70   ConvertToUTF16WithSubstitutions(bytes, codepage, utf16);
  71 }
  72
  73 }  // namespace
  74
  75 ProxyScriptFetcherImpl::ProxyScriptFetcherImpl(
  76     URLRequestContext* url_request_context)
  77     : url_request_context_(url_request_context),
  78       buf_(new IOBuffer(kBufSize)),
  79       next_id_(0),
  80       cur_request_id_(0),
  81       result_code_(OK),
  82       result_text_(NULL),
  83       max_response_bytes_(kDefaultMaxResponseBytes),
  84       max_duration_(base::TimeDelta::FromMilliseconds(kDefaultMaxDurationMs)),
  85       weak_factory_(this) {
  86   DCHECK(url_request_context);
  87 }
  88
  89 ProxyScriptFetcherImpl::~ProxyScriptFetcherImpl() {
  90   // The URLRequest's destructor will cancel the outstanding request, and
  91   // ensure that the delegate (this) is not called again.
  92 }
  93
  94 base::TimeDelta ProxyScriptFetcherImpl::SetTimeoutConstraint(
  95     base::TimeDelta timeout) {
  96   base::TimeDelta prev = max_duration_;
  97   max_duration_ = timeout;
  98   return prev;
  99 }
 100
 101 size_t ProxyScriptFetcherImpl::SetSizeConstraint(size_t size_bytes) {
 102   size_t prev = max_response_bytes_;
 103   max_response_bytes_ = size_bytes;
 104   return prev;
 105 }
 106
 107 void ProxyScriptFetcherImpl::OnResponseCompleted(URLRequest* request) {
 108   DCHECK_EQ(request, cur_request_.get());
 109
 110   // Use |result_code_| as the request's error if we have already set it to
 111   // something specific.
 112   if (result_code_ == OK && !request->status().is_success())
 113     result_code_ = request->status().error();
 114
 115   FetchCompleted();
 116 }
 117
 118 int ProxyScriptFetcherImpl::Fetch(
 119     const GURL& url, base::string16* text, const CompletionCallback& callback) {
 120   // It is invalid to call Fetch() while a request is already in progress.
 121   DCHECK(!cur_request_.get());
 122   DCHECK(!callback.is_null());
 123   DCHECK(text);
 124
 125   // Handle base-64 encoded data-urls that contain custom PAC scripts.
 126   if (url.SchemeIs("data")) {
 127     std::string mime_type;
 128     std::string charset;
 129     std::string data;
 130     if (!DataURL::Parse(url, &mime_type, &charset, &data))
 131       return ERR_FAILED;
 132
 133     ConvertResponseToUTF16(charset, data, text);
 134     return OK;
 135   }
 136
 137   DCHECK(fetch_start_time_.is_null());
 138   fetch_start_time_ = base::Time::Now();
 139
 140   cur_request_ =
 141       url_request_context_->CreateRequest(url, DEFAULT_PRIORITY, this);
 142   cur_request_->set_method("GET");
 143
 144   // Make sure that the PAC script is downloaded using a direct connection,
 145   // to avoid circular dependencies (fetching is a part of proxy resolution).
 146   // Also disable the use of the disk cache. The cache is disabled so that if
 147   // the user switches networks we don't potentially use the cached response
 148   // from old network when we should in fact be re-fetching on the new network.
 149   // If the PAC script is hosted on an HTTPS server we bypass revocation
 150   // checking in order to avoid a circular dependency when attempting to fetch
 151   // the OCSP response or CRL. We could make the revocation check go direct but
 152   // the proxy might be the only way to the outside world.
 153   cur_request_->SetLoadFlags(LOAD_BYPASS_PROXY | LOAD_DISABLE_CACHE |
 154                              LOAD_DISABLE_CERT_REVOCATION_CHECKING);
 155
 156   // Save the caller's info for notification on completion.
 157   callback_ = callback;
 158   result_text_ = text;
 159
 160   bytes_read_so_far_.clear();
 161
 162   // Post a task to timeout this request if it takes too long.
 163   cur_request_id_ = ++next_id_;
 164
 165   base::ThreadTaskRunnerHandle::Get()->PostDelayedTask(
 166       FROM_HERE, base::Bind(&ProxyScriptFetcherImpl::OnTimeout,
 167                             weak_factory_.GetWeakPtr(), cur_request_id_),
 168       max_duration_);
 169
 170   // Start the request.
 171   cur_request_->Start();
 172   return ERR_IO_PENDING;
 173 }
 174
 175 void ProxyScriptFetcherImpl::Cancel() {
 176   // ResetCurRequestState will free the URLRequest, which will cause
 177   // cancellation.
 178   ResetCurRequestState();
 179 }
 180
 181 URLRequestContext* ProxyScriptFetcherImpl::GetRequestContext() const {
 182   return url_request_context_;
 183 }
 184
 185 void ProxyScriptFetcherImpl::OnAuthRequired(URLRequest* request,
 186                                             AuthChallengeInfo* auth_info) {
 187   DCHECK_EQ(request, cur_request_.get());
 188   // TODO(eroman): http://crbug.com/77366
 189   LOG(WARNING) << "Auth required to fetch PAC script, aborting.";
 190   result_code_ = ERR_NOT_IMPLEMENTED;
 191   request->CancelAuth();
 192 }
 193
 194 void ProxyScriptFetcherImpl::OnSSLCertificateError(URLRequest* request,
 195                                                    const SSLInfo& ssl_info,
 196                                                    bool fatal) {
 197   DCHECK_EQ(request, cur_request_.get());
 198   // Revocation check failures are not fatal.
 199   if (IsCertStatusMinorError(ssl_info.cert_status)) {
 200     request->ContinueDespiteLastError();
 201     return;
 202   }
 203   LOG(WARNING) << "SSL certificate error when fetching PAC script, aborting.";
 204   // Certificate errors are in same space as net errors.
 205   result_code_ = MapCertStatusToNetError(ssl_info.cert_status);
 206   request->Cancel();
 207 }
 208
 209 void ProxyScriptFetcherImpl::OnResponseStarted(URLRequest* request) {
 210   DCHECK_EQ(request, cur_request_.get());
 211
 212   if (!request->status().is_success()) {
 213     OnResponseCompleted(request);
 214     return;
 215   }
 216
 217   // Require HTTP responses to have a success status code.
 218   if (request->url().SchemeIsHTTPOrHTTPS()) {
 219     // NOTE about status codes: We are like Firefox 3 in this respect.
 220     // {IE 7, Safari 3, Opera 9.5} do not care about the status code.
 221     if (request->GetResponseCode() != 200) {
 222       VLOG(1) << "Fetched PAC script had (bad) status line: "
 223               << request->response_headers()->GetStatusLine();
 224       result_code_ = ERR_PAC_STATUS_NOT_OK;
 225       request->Cancel();
 226       return;
 227     }
 228
 229     // NOTE about mime types: We do not enforce mime types on PAC files.
 230     // This is for compatibility with {IE 7, Firefox 3, Opera 9.5}. We will
 231     // however log mismatches to help with debugging.
 232     std::string mime_type;
 233     cur_request_->GetMimeType(&mime_type);
 234     if (!IsPacMimeType(mime_type)) {
 235       VLOG(1) << "Fetched PAC script does not have a proper mime type: "
 236               << mime_type;
 237     }
 238   }
 239
 240   ReadBody(request);
 241 }
 242
 243 void ProxyScriptFetcherImpl::OnReadCompleted(URLRequest* request,
 244                                              int num_bytes) {
 245   DCHECK_EQ(request, cur_request_.get());
 246   if (ConsumeBytesRead(request, num_bytes)) {
 247     // Keep reading.
 248     ReadBody(request);
 249   }
 250 }
 251
 252 void ProxyScriptFetcherImpl::ReadBody(URLRequest* request) {
 253   // Read as many bytes as are available synchronously.
 254   while (true) {
 255     int num_bytes;
 256     if (!request->Read(buf_.get(), kBufSize, &num_bytes)) {
 257       // Check whether the read failed synchronously.
 258       if (!request->status().is_io_pending())
 259         OnResponseCompleted(request);
 260       return;
 261     }
 262     if (!ConsumeBytesRead(request, num_bytes))
 263       return;
 264   }
 265 }
 266
 267 bool ProxyScriptFetcherImpl::ConsumeBytesRead(URLRequest* request,
 268                                               int num_bytes) {
 269   if (num_bytes <= 0) {
 270     // Error while reading, or EOF.
 271     OnResponseCompleted(request);
 272     return false;
 273   }
 274
 275   // Enforce maximum size bound.
 276   if (num_bytes + bytes_read_so_far_.size() >
 277       static_cast<size_t>(max_response_bytes_)) {
 278     result_code_ = ERR_FILE_TOO_BIG;
 279     request->Cancel();
 280     return false;
 281   }
 282
 283   bytes_read_so_far_.append(buf_->data(), num_bytes);
 284   return true;
 285 }
 286
 287 void ProxyScriptFetcherImpl::FetchCompleted() {
 288   if (result_code_ == OK) {
 289     // Calculate duration of time for proxy script fetch to complete.
 290     DCHECK(!fetch_start_time_.is_null());
 291     UMA_HISTOGRAM_TIMES("Net.ProxyScriptFetcher.SuccessDuration",
 292                         base::Time::Now() - fetch_start_time_);
 293
 294     // The caller expects the response to be encoded as UTF16.
 295     std::string charset;
 296     cur_request_->GetCharset(&charset);
 297     ConvertResponseToUTF16(charset, bytes_read_so_far_, result_text_);
 298   } else {
 299     // On error, the caller expects empty string for bytes.
 300     result_text_->clear();
 301   }
 302
 303   int result_code = result_code_;
 304   CompletionCallback callback = callback_;
 305
 306   ResetCurRequestState();
 307
 308   callback.Run(result_code);
 309 }
 310
 311 void ProxyScriptFetcherImpl::ResetCurRequestState() {
 312   cur_request_.reset();
 313   cur_request_id_ = 0;
 314   callback_.Reset();
 315   result_code_ = OK;
 316   result_text_ = NULL;
 317   fetch_start_time_ = base::Time();
 318 }
 319
 320 void ProxyScriptFetcherImpl::OnTimeout(int id) {
 321   // Timeout tasks may outlive the URLRequest they reference. Make sure it
 322   // is still applicable.
 323   if (cur_request_id_ != id)
 324     return;
 325
 326   DCHECK(cur_request_.get());
 327   result_code_ = ERR_TIMED_OUT;
 328   cur_request_->Cancel();
 329 }
 330
 331 }  // namespace net