Failed to load PDF in chrome/Firefox/IE

笔者的公司搭建了一个Nexus服务器,用来管理我们自己的项目Release构件和Site文档.

今天的问题是当用户访问一个Site里的PDF文件的时候，报错说“detected that the network has been closed.”

但是用右键点击，然后另存为，却每次都能成功下载下来.本来这几天公司的网络确实是不稳定，刚开始还真的以为是网络问题.

后来仔细观察了http请求后发现，每次报错后浏览器都还在不停的发送 http range 请求. 我们用的是Adbobe的PDF插件.

• Chrome问题

Chrome里有一个自带的plugin，可以通过 chrome://plugins 进入plugin配置页面，找到自带的Chrome PDF viewer.

这个Chrome插件发现提供PDF的Nexus服务器支持range请求后，在读取完response的header后会主动abort response.

然后改为通过XHR分批发送Http Range 请求，此时Adbobe的PDF插件就报network已经关闭了。这里的主要原因是两个插件互相冲突，

而且chrome自带的PDF viewer有bug，不能正确组装返回的range response. 解决办法可以是禁用Chrome PDF viewer。

• Firefox问题/IE问题

老版本的IE和Firefox都没问题，有问题的是IE10，Firefox25. 原因是老版本的IE/Firefox里的adobe插件不会发生http range 请求.

新版的IE/Firefox发生了range请求后，服务器也正常的给了response，结果可笑的是adobe 插件不能正确组装，悲剧。

最后想到的就是禁用range请求.但是在流浪器上是做不到的，因为是插件自己发送的.而Nexus服务器2.7.1是hard code支持这个特性的。

range request是http1.1标准里的特性，支持是理所应当的。 下面是NexusContextServlet.Java的部分代码.红色部分显示了这个特性。

protected void doGetFile(final HttpServletRequest request, final HttpServletResponse response,
      final StorageFileItem file) throws IOException
  {
    // ETag, in "shaved" form of {SHA1{e5c244520e897865709c730433f8b0c44ef271f1}} (without quotes)
    // or null if file does not have SHA1 (like Virtual) or generated items (as their SHA1 would correspond to template,
    // not to actual generated content).
    final String etag;
    if (!file.isContentGenerated() && !file.isVirtual()
        && file.getRepositoryItemAttributes().containsKey(StorageFileItem.DIGEST_SHA1_KEY)) {
      etag = "{SHA1{" + file.getRepositoryItemAttributes().get(StorageFileItem.DIGEST_SHA1_KEY) + "}}";
      // tag header ETag: "{SHA1{e5c244520e897865709c730433f8b0c44ef271f1}}", quotes are must by RFC
      response.setHeader("ETag", "\"" + etag + "\"");
    }
    else {
      etag = null;
    }
    // content-type
    response.setHeader("Content-Type", file.getMimeType());

    // last-modified
    response.setDateHeader("Last-Modified", file.getModified());

    // content-length, if known
    if (file.getLength() != ContentLocator.UNKNOWN_LENGTH) {
      // Note: response.setContentLength Servlet API method uses ints (max 2GB file)!
      // TODO: apparently, some Servlet containers follow serlvet API and assume
      // contents can have 2GB max, so even this workaround below in inherently unsafe.
      // Jetty is checked, and supports this (uses long internally), but unsure for other containers
      response.setHeader("Content-Length", String.valueOf(file.getLength()));
    }

    // handle conditional GETs only for "static" content, actual content stored, not generated
    if (!file.isContentGenerated() && file.getResourceStoreRequest().getIfModifiedSince() != 0
        && file.getModified() <= file.getResourceStoreRequest().getIfModifiedSince()) {
      // this is a conditional GET using time-stamp
      response.setStatus(HttpServletResponse.SC_NOT_MODIFIED);
    }
    else if (!file.isContentGenerated() && file.getResourceStoreRequest().getIfNoneMatch() != null && etag != null
        && file.getResourceStoreRequest().getIfNoneMatch().equals(etag)) {
      // this is a conditional GET using ETag
      response.setStatus(HttpServletResponse.SC_NOT_MODIFIED);
    }
    else {
      // NEXUS-5023 disable IE for sniffing into response content
      response.setHeader("X-Content-Type-Options", "nosniff");

      final List<Range<Long>> ranges = getRequestedRanges(request, file.getLength());

      // pour the content, but only if needed (this method will be called even for HEAD reqs, but with content tossed
      // away), so be conservative as getting input stream involves locking etc, is expensive
      final boolean contentNeeded = "GET".equalsIgnoreCase(request.getMethod());
      if (ranges.isEmpty()) {
        if (contentNeeded) {
          try (final InputStream in = file.getInputStream()) {
            sendContent(in, response);
          }
        }
      }
      else if (ranges.size() > 1) {
        response.setStatus(HttpServletResponse.SC_NOT_IMPLEMENTED);
        renderer.renderErrorPage(request, response, file.getResourceStoreRequest(), new UnsupportedOperationException(
            "Multiple ranges not yet supported!"));
      }
      else {
        final Range<Long> range = ranges.get(0);
        if (!isRequestedRangeSatisfiable(file, range)) {
          response.setStatus(HttpServletResponse.SC_REQUESTED_RANGE_NOT_SATISFIABLE);
          response.setHeader("Content-Length", "0");
          response.setHeader("Content-Range", "bytes */" + file.getLength());
          return;
        }
        final long bodySize = range.upperEndpoint() - range.lowerEndpoint();
        response.setStatus(HttpServletResponse.SC_PARTIAL_CONTENT);
        response.setHeader("Content-Length", String.valueOf(bodySize));
        response.setHeader("Content-Range",
            range.lowerEndpoint() + "-" + range.upperEndpoint() + "/" + file.getLength());
        if (contentNeeded) {
          try (final InputStream in = file.getInputStream()) {
            in.skip(range.lowerEndpoint());
            sendContent(limit(in, bodySize), response);
          }
        }
      }
    }
  }

最后幸好我们的请求在到Nexus之前有一个Apache proxy,根据http1.1标准，range request必须是client先发送range 请求，如果服务器支持

那么才返回状态嘛206和部分数据。所以最后我在proxy中把range header 强行去掉。这样服务器就不会返回206，而是返回200了.

也没有content-range header返回了，所以插件就会按照老实的方式，等待下载完成，然后渲染打开PDF文件.

最后的解决方案就是在Apache中增加如下配置.

LoadModule headers_module modules/mod_headers.so

RequestHeader unset Range