[NaCl SDK] Add HTTP mount.

native_client_sdk/src/libraries/nacl_mounts/mount_http.cc

+/* Copyright (c) 2012 The Chromium Authors. All rights reserved.
+ * Use of this source code is governed by a BSD-style license that can be
+ * found in the LICENSE file.
+ */
+
+#include "nacl_mounts/mount_http.h"
+#include <assert.h>
+#include <ctype.h>
+#include <errno.h>
+#include <ppapi/c/pp_errors.h>
+#include <stdio.h>
+#include <string.h>
+#include <vector>
+#include "utils/auto_lock.h"
+
+namespace {
+
+static const int32_t STATUSCODE_OK = 200;
+static const int32_t STATUSCODE_PARTIAL_CONTENT = 206;
+
+std::string NormalizeHeaderKey(const std::string& s) {
+  // Capitalize the first letter and any letter following a hyphen:
+  // e.g. ACCEPT-ENCODING -> Accept-Encoding
+  std::string result;
+  bool upper = true;
+  for (size_t i = 0; i < s.length(); ++i) {
+    char c = s[i];
+    result += upper ? toupper(c) : tolower(c);
+    upper = c == '-';
+  }
+
+  return result;
+}
+
+StringMap_t ParseHeaders(const char* headers, int32_t headers_length) {
+  enum State {
+    FINDING_KEY,
+    SKIPPING_WHITESPACE,
+    FINDING_VALUE,
+  };
+
+  StringMap_t result;
+  std::string key;
+  std::string value;
+
+  State state;

[noelallen1, 2013/01/14 23:53:41]

Uninitialized?

[binji, 2013/01/15 01:42:49]

Done.

+  const char* start = headers;
+  for (int i = 0; i < headers_length; ++i) {
+    switch (state) {
+      case FINDING_KEY:
+        if (headers[i] == ':') {
+          // Found key.
+          key.assign(start, &headers[i] - start);
+          key = NormalizeHeaderKey(key);
+          state = SKIPPING_WHITESPACE;
+        }
+        break;
+
+      case SKIPPING_WHITESPACE:
+        if (headers[i] == ' ') {
+          // Found whitespace, keep going...
+          break;
+        }
+
+        // Found a non-whitespace, mark this as the start of the value.
+        start = &headers[i + 1];
+        state = FINDING_VALUE;
+        // Fallthrough to start processing value without incrementing i.
+
+      case FINDING_VALUE:
+        if (headers[i] == '\n') {
+          // Found value.
+          value.assign(start, &headers[i] - start);
+          result[key] = value;
+
+          start = &headers[i + 1];
+          state = FINDING_KEY;
+        }

[noelallen1, 2013/01/14 23:53:41]

Can we assume no whitespace after \n?  Such as \t ' ' or a second \n?

[binji, 2013/01/15 01:42:49]

Looks like trailing whitespace MAY be removed, but is not required. Probably
should handle it here though.

http://www.w3.org/Protocols/rfc2616/rfc2616-sec4.html

+        break;
+    }
+  }
+
+  return result;
+}
+
+bool ParseContentLength(const StringMap_t& headers, size_t* content_length) {
+  StringMap_t::const_iterator iter = headers.find("Content-Length");
+  if (iter == headers.end())
+    return false;
+
+  *content_length = strtoul(iter->second.c_str(), NULL, 10);
+  return true;
+}
+
+bool ParseContentRange(const StringMap_t& headers, size_t* read_start,
+                       size_t* read_end, size_t* entity_length) {
+  StringMap_t::const_iterator iter = headers.find("Content-Range");
+  if (iter == headers.end())
+    return false;
+
+  // The key should look like "bytes %d-%d/%d" or "bytes %d-%d/*". The last
+  // value is the entity length, which can potentially be * (i.e. unknown).
+  int read_start_int;
+  int read_end_int;
+  int entity_length_int;
+  int result = sscanf(iter->second.c_str(), "bytes %d-%d/%d", &read_start_int,
+                      &read_end_int, &entity_length_int);
+
+  if (result == 2) {
+    *read_start = read_start_int;
+    *read_end = read_end_int;
+    *entity_length = 0;
+    return true;
+  } else if (result == 3) {
+    *read_start = read_start_int;
+    *read_end = read_end_int;
+    *entity_length = entity_length_int;
+    return true;
+  }
+
+  return false;
+}
+
+class MountNodeHttp : public MountNode {
+ public:
+  virtual int FSync();
+  virtual int GetDents(size_t offs, struct dirent* pdir, size_t count);
+  virtual int GetStat(struct stat* stat);
+  virtual int Read(size_t offs, void* buf, size_t count);
+  virtual int Truncate(size_t size);
+  virtual int Write(size_t offs, const void* buf, size_t count);
+  virtual size_t GetSize();
+
+ protected:
+  MountNodeHttp(Mount* mount, int ino, int dev, const std::string& url);
+  virtual bool Init(int mode, short uid, short gid);
+  virtual int Close();
+
+ private:
+  bool OpenUrl(const char* method,
+               StringMap_t* request_headers,
+               PP_Resource* out_loader,
+               PP_Resource* out_request,
+               PP_Resource* out_response,
+               int32_t* out_statuscode,
+               StringMap_t* out_response_headers);
+
+  std::string url_;
+  std::vector<char> buffer_;

[noelallen1, 2013/01/14 23:53:41]

When does buffer_ get sized?

[binji, 2013/01/15 01:42:49]

Done.

+
+  friend class ::MountHttp;
+};
+
+int MountNodeHttp::FSync() {
+  errno = ENOSYS;
+  return -1;
+}
+
+int MountNodeHttp::GetDents(size_t offs, struct dirent* pdir, size_t count) {
+  errno = ENOSYS;
+  return -1;
+}
+
+int MountNodeHttp::GetStat(struct stat* stat) {
+  AutoLock lock(&lock_);
+
+  StringMap_t headers;
+  PP_Resource loader;
+  PP_Resource request;
+  PP_Resource response;
+  int32_t statuscode;
+  StringMap_t response_headers;
+  if (!OpenUrl("HEAD", &headers, &loader, &request, &response, &statuscode,
+               &response_headers)) {
+    // errno is already set by OpenUrl.
+    return -1;
+  }
+
+  ScopedResource scoped_loader(mount_->ppapi(), loader);
+  ScopedResource scoped_request(mount_->ppapi(), request);
+  ScopedResource scoped_response(mount_->ppapi(), response);
+
+  // Fill in known info here.
+  memcpy(stat, &stat_, sizeof(stat_));
+
+  size_t entity_length;
+  if (ParseContentLength(response_headers, &entity_length))
+    stat->st_size = static_cast<off_t>(entity_length);
+  else
+    stat->st_size = 0;
+
+  stat->st_atime = 0; // TODO(binji): Use "Last-Modified".
+  stat->st_mtime = 0;
+  stat->st_ctime = 0;
+
+  return 0;
+}
+
+int MountNodeHttp::Read(size_t offs, void* buf, size_t count) {
+  AutoLock lock(&lock_);
+  StringMap_t headers;
+
+  char buffer[100];

[noelallen1, 2013/01/14 23:53:41]

100 bytes is a random size.

[binji, 2013/01/15 01:42:49]

It just needs to be big enough to fit the string below. Should I add a comment?

+  // Range request is inclusive: 0-99 returns 100 bytes.
+  snprintf(&buffer[0], 100, "bytes=%d-%d", offs, offs + count - 1);

[noelallen1, 2013/01/14 23:53:41]

%lu?

[binji, 2013/01/15 01:42:49]

Looks like the Google C++ style guide suggests PRIuS
(http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml?showone=64-bit...)
-- since I have to support Windows as well here, I added a new header to
determine the correct format string.

+  headers["Range"] = buffer;
+
+  PP_Resource loader;
+  PP_Resource request;
+  PP_Resource response;
+  int32_t statuscode;
+  StringMap_t response_headers;
+  if (!OpenUrl("GET", &headers, &loader, &request, &response, &statuscode,
+               &response_headers)) {
+    // errno is already set by OpenUrl.
+    return 0;
+  }
+
+  PepperInterface* ppapi = mount_->ppapi();
+  ScopedResource scoped_loader(ppapi, loader);
+  ScopedResource scoped_request(ppapi, request);
+  ScopedResource scoped_response(ppapi, response);
+
+  size_t read_start = 0;
+  if (statuscode == STATUSCODE_OK) {
+    // No partial result, read everything starting from the part we care about.
+    size_t content_length;
+    if (ParseContentLength(response_headers, &content_length)) {
+      if (offs >= content_length) {
+        errno = EINVAL;
+        return 0;
+      }
+
+      // Clamp count, if trying to read past the end of the file.
+      if (offs + count > content_length) {
+        count = content_length - offs;
+      }
+    }
+  } else if (statuscode == STATUSCODE_PARTIAL_CONTENT) {
+    // Determine from the headers where we are reading.
+    size_t read_end;
+    size_t entity_length;
+    if (ParseContentRange(response_headers, &read_start, &read_end,
+                          &entity_length)) {
+      if (read_start > offs || read_start > read_end) {
+        // Shouldn't happen.
+        errno = EINVAL;
+        return 0;
+      }
+
+      // Clamp count, if trying to read past the end of the file.
+      count = std::min(read_end - read_start, count);
+    } else {
+      // Partial Content without Content-Range. Assume that the server gave us
+      // exactly what we asked for. This can happen even when the server
+      // returns 200 -- the cache may return 206 in this case, but not modify
+      // the headers.
+      read_start = offs;
+    }
+  }
+
+  URLLoaderInterface* loader_interface = ppapi->GetURLLoaderInterface();
+
+  size_t bytes_to_read;
+  int32_t bytes_read;
+  while (read_start < offs) {
+    // We aren't yet at the location where we want to start reading. Read into
+    // our dummy buffer until then.
+    bytes_to_read = std::min(offs - read_start, buffer_.size());
+    bytes_read = loader_interface->ReadResponseBody(
+        loader, buffer_.data(), bytes_to_read, PP_BlockUntilComplete());
+
+    if (bytes_read < 0) {
+      errno = PPErrorToErrno(bytes_read);
+      return 0;
+    }
+
+    assert(bytes_read <= bytes_to_read);
+    read_start += bytes_read;
+  }
+
+  // At the read start, now we can read into the correct buffer.
+  char* out_buffer = static_cast<char*>(buf);
+  bytes_to_read = count;
+  while (bytes_to_read > 0) {
+    bytes_read = loader_interface->ReadResponseBody(
+        loader, out_buffer, bytes_to_read, PP_BlockUntilComplete());
+
+    if (bytes_read == 0) {
+      // This is not an error -- it may just be that we were trying to read
+      // more data than exists.
+      return count - bytes_to_read;
+    }
+
+    if (bytes_read < 0) {
+      errno = PPErrorToErrno(bytes_read);
+      return count - bytes_to_read;
+    }
+
+    assert(bytes_read <= bytes_to_read);
+    bytes_to_read -= bytes_read;
+    out_buffer += bytes_read;
+  }
+
+  return count;
+}
+
+int MountNodeHttp::Truncate(size_t size) {
+  errno = ENOSYS;
+  return -1;
+}
+
+int MountNodeHttp::Write(size_t offs, const void* buf, size_t count) {
+  // TODO(binji): supprt POST?
+  errno = ENOSYS;
+  return -1;
+}
+
+size_t MountNodeHttp::GetSize() {
+  return 0;

[noelallen1, 2013/01/14 23:53:41]

Isn't this in a response header somewhere?

[binji, 2013/01/15 01:42:49]

Done.

+}
+
+MountNodeHttp::MountNodeHttp(Mount* mount, int ino, int dev,
+                           const std::string& url)
+    : MountNode(mount, ino, dev),
+      url_(url) {
+}
+
+bool MountNodeHttp::Init(int mode, short uid, short gid) {
+  return MountNode::Init(mode, uid, gid);

[noelallen1, 2013/01/14 23:53:41]

This always returns true...  Which means an open of a non-existent URL will
appear to be a failed Read.

[binji, 2013/01/15 01:42:49]

I've changed this to do an initial stat (which does a HEAD request, should be
quick) to determine if the file exists.

+}
+
+int MountNodeHttp::Close() {
+  return 0;
+}
+
+bool MountNodeHttp::OpenUrl(const char* method,
+                           StringMap_t* request_headers,
+                           PP_Resource* out_loader,
+                           PP_Resource* out_request,
+                           PP_Resource* out_response,
+                           int32_t* out_statuscode,
+                           StringMap_t* out_response_headers) {
+  // Assume lock_ is already held.
+
+  PepperInterface* ppapi = mount_->ppapi();
+
+  MountHttp* mount_http = static_cast<MountHttp*>(mount_);
+  ScopedResource request(ppapi,
+                         mount_http->MakeUrlRequestInfo(url_, method,
+                                                        request_headers));
+  if (!request.pp_resource()) {
+    errno = EINVAL;
+    return false;
+  }
+
+  URLLoaderInterface* loader_interface = ppapi->GetURLLoaderInterface();
+  URLResponseInfoInterface* response_interface =
+      ppapi->GetURLResponseInfoInterface();
+  VarInterface* var_interface = ppapi->GetVarInterface();
+
+  ScopedResource loader(ppapi, loader_interface->Create(ppapi->GetInstance()));
+  if (!loader.pp_resource()) {
+    errno = EINVAL;
+    return false;
+  }
+
+  int32_t result = loader_interface->Open(
+      loader.pp_resource(), request.pp_resource(), PP_BlockUntilComplete());
+  if (result != PP_OK) {
+    errno = PPErrorToErrno(result);
+    return false;
+  }
+
+  ScopedResource response(
+      ppapi,
+      loader_interface->GetResponseInfo(loader.pp_resource()));
+  if (!response.pp_resource()) {
+    errno = EINVAL;
+    return false;
+  }
+
+  // Get response statuscode.
+  PP_Var statuscode = response_interface->GetProperty(
+      response.pp_resource(),
+      PP_URLRESPONSEPROPERTY_STATUSCODE);
+
+  if (statuscode.type != PP_VARTYPE_INT32) {
+    errno = EINVAL;
+    return false;
+  }
+
+  *out_statuscode = statuscode.value.as_int;
+
+  // Only accept OK or Partial Content.
+  if (*out_statuscode != STATUSCODE_OK &&
+      *out_statuscode != STATUSCODE_PARTIAL_CONTENT) {
+    errno = EINVAL;
+    return false;
+  }
+
+  // Get response headers.
+  PP_Var response_headers_var = response_interface->GetProperty(
+      response.pp_resource(),
+      PP_URLRESPONSEPROPERTY_HEADERS);
+
+  uint32_t response_headers_length;
+  const char* response_headers_str = var_interface->VarToUtf8(
+      response_headers_var,
+      &response_headers_length);
+
+  *out_loader = loader.Release();
+  *out_request = request.Release();
+  *out_response = response.Release();
+  *out_response_headers = ParseHeaders(response_headers_str,
+                                       response_headers_length);
+
+  return true;
+}
+
+
+}  // namespace
+
+MountNode *MountHttp::Open(const Path& path, int mode) {
+  assert(url_root_.empty() || url_root_[url_root_.length() - 1] == '/');
+
+  std::string url = url_root_ + (path.IsAbsolute() ?
+      path.Range(1, path.Size()) :
+      path.Join());
+
+  const int ino = 1;
+  const int USR_ID = 1001;
+  const int GRP_ID = 1002;
+  MountNodeHttp* node = new MountNodeHttp(this, ino, dev_, url);
+  if (!node->Init(mode, USR_ID, GRP_ID)) {
+    node->Release();
+    return NULL;
+  }
+
+  return node;
+}
+
+int MountHttp::Close(MountNode* node) {
+  AutoLock lock(&lock_);
+  node->Close();
+  node->Release();
+  return 0;
+}
+
+int MountHttp::Unlink(const Path& path) {
+  errno = ENOSYS;
+  return -1;
+}
+
+int MountHttp::Mkdir(const Path& path, int permissions) {
+  errno = ENOSYS;
+  return -1;
+}
+
+int MountHttp::Rmdir(const Path& path) {
+  errno = ENOSYS;
+  return -1;
+}
+
+int MountHttp::Remove(const Path& path) {
+  errno = ENOSYS;
+  return -1;
+}
+
+PP_Resource MountHttp::MakeUrlRequestInfo(
+    const std::string& url,
+    const char* method,
+    StringMap_t* additional_headers) {
+  URLRequestInfoInterface* interface = ppapi_->GetURLRequestInfoInterface();
+  VarInterface* var_interface = ppapi_->GetVarInterface();
+
+  PP_Resource request_info = interface->Create(ppapi_->GetInstance());
+  if (!request_info)
+    return 0;
+
+  interface->SetProperty(
+      request_info, PP_URLREQUESTPROPERTY_URL,
+      var_interface->VarFromUtf8(url.c_str(), url.length()));
+  interface->SetProperty(request_info, PP_URLREQUESTPROPERTY_METHOD,
+                         var_interface->VarFromUtf8(method, strlen(method)));
+  interface->SetProperty(request_info,
+                         PP_URLREQUESTPROPERTY_ALLOWCROSSORIGINREQUESTS,
+                         PP_MakeBool(allow_cors_ ? PP_TRUE : PP_FALSE));
+  interface->SetProperty(request_info, PP_URLREQUESTPROPERTY_ALLOWCREDENTIALS,
+                         PP_MakeBool(allow_credentials_ ? PP_TRUE : PP_FALSE));
+
+  // Merge the mount headers with the request headers. If the field is already
+  // set it |additional_headers|, don't use the one from headers_.
+  for (StringMap_t::iterator iter = headers_.begin(); iter != headers_.end();
+      ++iter) {
+    const std::string& key = NormalizeHeaderKey(iter->first);
+    if (additional_headers->find(key) == additional_headers->end()) {
+      additional_headers->insert(std::make_pair(key, iter->second));
+    }
+  }
+
+  // Join the headers into one string.
+  std::string headers;
+  for (StringMap_t::iterator iter = additional_headers->begin();
+      iter != additional_headers->end(); ++iter) {
+    headers += iter->first + ": " + iter->second + '\n';
+  }
+
+  interface->SetProperty(
+      request_info, PP_URLREQUESTPROPERTY_HEADERS,
+      var_interface->VarFromUtf8(headers.c_str(), headers.length()));
+
+  return request_info;
+}
+
+MountHttp::MountHttp()
+    : allow_cors_(false),
+      allow_credentials_(false) {
+}
+
+bool MountHttp::Init(int dev, StringMap_t& args, PepperInterface* ppapi) {
+  if (!Mount::Init(dev, args, ppapi))
+    return false;
+
+  // Parse mount args.
+  for (StringMap_t::iterator iter = args.begin(); iter != args.end(); ++iter) {
+    if (iter->first == "SOURCE") {
+      url_root_ = iter->second;
+
+      // Make sure url_root_ ends with a slash.
+      if (!url_root_.empty() && url_root_[url_root_.length() - 1] != '/') {
+        url_root_ += '/';
+      }
+    } else if (iter->first == "allow_cross_origin_requests") {
+      allow_cors_ = iter->second == "true";
+    } else if (iter->first == "allow_credentials") {
+      allow_credentials_ = iter->second == "true";
+    } else {
+      // Assume it is a header to pass to an HTTP request.
+      headers_[NormalizeHeaderKey(iter->first)] = iter->second;
+    }
+  }
+
+  return true;
+}
+
+void MountHttp::Destroy() {
+}