Add read-only access to POST data for webRequest's onBeforeRequest

chrome/browser/extensions/api/web_request/form_data_parser.cc

Index: chrome/browser/extensions/api/web_request/form_data_parser.cc
diff --git a/chrome/browser/extensions/api/web_request/form_data_parser.cc b/chrome/browser/extensions/api/web_request/form_data_parser.cc
new file mode 100644
index 0000000000000000000000000000000000000000..0ea100cf856a4919b653c06aabd20150b98d95de
--- /dev/null
+++ b/chrome/browser/extensions/api/web_request/form_data_parser.cc
@@ -0,0 +1,730 @@
+// 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 "chrome/browser/extensions/api/web_request/form_data_parser.h"
+
+#include <vector>
+
+#include "base/string_util.h"
+#include "base/values.h"
+#include "net/base/escape.h"
+#include "net/url_request/url_request.h"
+
+using base::DictionaryValue;
+using base::ListValue;
+using base::StringPiece;
+
+namespace extensions {
+
+// Parses URLencoded forms, see
+// http://www.w3.org/TR/REC-html40-971218/interact/forms.html#h-17.13.4.1 .
+class FormDataParserUrlEncoded : public FormDataParser {
+ public:
+  FormDataParserUrlEncoded();
+  virtual ~FormDataParserUrlEncoded();
+
+  // Implementation of FormDataParser.
+  virtual bool AllDataReadOK() OVERRIDE;
+  virtual bool GetNextNameValue(Result* result) OVERRIDE;
+  virtual bool SetSource(const base::StringPiece& source) OVERRIDE;
+
+ private:
+  // Gets next char from |source_|, seeks, and does book-keeping of = and &.
+  // Returns false if end of |source_| was reached, otherwise true.
+  bool GetNextChar(char* c);
+  // Once called the parser gives up and claims any results so far invalid.
+  void Abort();
+
+  base::StringPiece source_;
+  const char* source_end_;
+  bool aborted_;
+
+  // Variables from this block are only to be written to by GetNextChar.
+  const char* offset_;  // Next char to be read.
+  size_t equality_signs_;  // How many '=' were read so far.
+  size_t amp_signs_;       // How many '&' were read so far.
+  bool expect_equality_;   // Is the next trailing sign '=' (as opposed to '&')?
+
+  DISALLOW_COPY_AND_ASSIGN(FormDataParserUrlEncoded);
+};
+
+// The following class, FormDataParserMultipart, parses forms encoded as
+// multipart, defined in RFCs 2388 (specific to forms), 2046 (multipart
+// encoding) and 822 (MIME-headers).

[tkent, 2012/08/27 07:09:17]

Please do not refer to RFC 822, which was obsoleted by RFC 2822, which was
obsoleted by RFC 5322, which is the latest standard.

[vabr (Chromium), 2012/08/29 19:57:07]

Done.
Thanks for making me aware of this.

+//
+// Implementation details
+//
+// The original grammar from RFC 2046 is this, "multipart-body" being the root
+// non-terminal:
+//
+// boundary := 0*69<bchars> bcharsnospace
+// bchars := bcharsnospace / " "
+// bcharsnospace := DIGIT / ALPHA / "'" / "(" / ")" / "+" / "_" / ","
+//                  / "-" / "." / "/" / ":" / "=" / "?"
+// dash-boundary := "--" boundary
+// multipart-body := [preamble CRLF]
+//                        dash-boundary transport-padding CRLF
+//                        body-part *encapsulation
+//                        close-delimiter transport-padding
+//                        [CRLF epilogue]
+// transport-padding := *LWSP-char
+// encapsulation := delimiter transport-padding CRLF body-part
+// delimiter := CRLF dash-boundary
+// close-delimiter := delimiter "--"
+// preamble := discard-text
+// epilogue := discard-text
+// discard-text := *(*text CRLF) *text
+// body-part := MIME-part-headers [CRLF *OCTET]
+// OCTET := <any 0-255 octet value>
+//
+// Here, CRLF, DIGIT, and ALPHA stand for "\r\n", '0'-'9' and the set of letters
+// of the English alphabet, respectively.
+// The non-terminal "text" is presumably just any text, excluding line breaks.
+// The non-terminal "LWSP-char" is not directly defined in the original grammar
+// but it means "linear whitespace", which is a space or a horizontal tab.
+// The non-terminal "MIME-part-headers" is not discussed in RFC 2046, but is in
+// English defined in RFC 822, and can be presented as follows:
+//
+// MIME-part-headers := *MIME-part-header
+// MIME-part-header := name ':' *(text / whitespace) linebreak
+// linebreak := '\r' / '\n' / CRLF
+// whitespace := LWSP-char / CRLF LWSP-char
+// name := namechar *namechar
+// namechar := <ASCII char between 33 and 126, excluding ':'>
+//
+// This sets of rules together compose a grammar, with the root non-terminal
+// "multipart-body". This grammer defines a regular language. Indeed, if the
+// non-terminals are ordered in this way:
+// namechar < name < CRLF < DIGIT < ALPHA < text < LWSP-char < whitespace <
+//     linebreak < MIME-part-header < MIME-part-headers < bcharsnospace <
+//     bchars < boundary < dash-boundary < delimiter < close-delimiter <
+//     discard-text < transport-padding < OCTET < body-part < encapsulation <
+//     multipart-body
+// then it is easy to verify that whenever A<B then no grammar rule with head
+// A contains B in the body. By induction on the above order, each non-terminal
+// defines a regular language: a non-terminal C is defined by a rule C := exp,
+// where "exp" is an expression composed from character constants, non-terminals
+// less than C, and the following closure operations of regular languages:
+// concatenation, union and Kleene-star. By induction, all the lesser
+// non-terminals represent regular languages, thus "exp" also represents a
+// regular language. In particular, the root non-terminal (and thus the grammar)
+// defines a regular language.
+//
+// The FormDataParseMultipart class uses a finite automaton to represent this
+// language. It is easiest to view it in an extended form, with longer words
+// allowed to label a single transition to keep the number of states is low.
+// Importand states have full-word names, unimportant states (allways with only
+// one incoming label) have names abbreviating the incoming label, possibly
+// with an index.
+//
+// Automaton for "multipart-body":
+// Initial state = Start
+// Final states = {End, IgnoreEpilogue}
+// Implicit state (when a transition is missing) = Error
+// Transition table ('*' is a label matching everything not matched by other
+// labels leaving the same state):
+// FROM                 LABEL                           TO
+// Start                dash-boundary                   DB1
+//                      CR                              CR1
+//                      *                               IgnorePreamble
+// CR1                  LF                              Start
+//                      *                               IgnorePreamble
+// IgnorePreamble       CR                              CR1
+//                      *                               IgnorePreamble
+// DB1                  LWSP-char                       DB1
+//                      CR                              CR2
+// CR2                  LF                              Part
+// Part                 <ASCII 33-126, excluding ':'>   Name
+//                      CR                              CR3
+// Name                 <ASCII 33-126, excluding ':'>   Name
+//                      ':'                             Colon
+// Colon                LF                              End1
+//                      CR                              End2
+//                      *                               Colon
+// End1                 CR                              CR3
+//                      <ASCII 33-126, excluding ':'>   Name
+// End2                 LF                              End3
+//                      CR                              CR3
+//                      <ASCII 33-126, excluding ':'>   Name
+// End3                 LWSP-char                       Colon
+//                      CR                              CR3
+//                      <ASCII 33-126, excluding ':'>   Name
+// CR3                  LF                              PreData
+// PreData              dash-boundary                   DB2
+//                      CR                              CR4
+//                      *                               Data
+// CR4                  LF                              Data2
+//                      *                               Data
+// Data                 CR                              CR4
+//                      *                               Data
+// Data2                dash-boundary                   DB2
+//                      *                               CR4
+// DB2                  LWSP-char                       DB1
+//                      CR                              CR2
+//                      '-'                             D
+// D                    '-'                             End
+// End                  LWSP-char                       End
+//                      CR                              CR5
+// CR5                  LF                              IgnoreEpilogue
+// IgnoreEpilogue       *                               IgnoreEpilogue
+//
+// The automaton itself only allows to check that the input is a well-formed
+// multipart encoding of a form. To also extract the data, additional logic is
+// added:
+//  * The header "Content-Disposition" (read between Part and PreData) contains
+//    the elements name=... and optionally filename=... The former is the name
+//    of the corresponding field of a form. The latter is only present if that
+//    field was a file-upload, and contains the path to the uploaded file.
+//  * The data of a message part is read between PreData and DB2, excluding the
+//    last CR LF dash-boundary.
+//
+// IMPORTANT NOTE
+// This parser supports multiple sources, i.e., SetSource can be called multiple
+// times if the input is spread over several byte blocks. However, the split
+// must not occur in the middle of a transition of the above described automata,
+// e.g., if there is a transition StateA --dash-boundary--> StateB, then the
+// whole string with the dash--boundary bust be contained in the first source,

[tkent, 2012/08/27 07:09:17]

bust -> must?

[vabr (Chromium), 2012/08/29 19:57:07]

Rewritten in the meantime.

+// or in the other. Also, the split must not occur in the middle of a header,
+// or a part body data. A message part from one source must be read via
+// GetNextNameValue before setting up a new source.
+class FormDataParserMultipart : public FormDataParser {
+ public:
+  explicit FormDataParserMultipart(const std::string& boundary_separator);
+  virtual ~FormDataParserMultipart();
+
+  // Implementation of FormDataParser.
+  virtual bool AllDataReadOK() OVERRIDE;
+  virtual bool GetNextNameValue(Result* result) OVERRIDE;
+  virtual bool SetSource(const base::StringPiece& source) OVERRIDE;
+
+ private:
+  // State and Transition are numbered to make sure they form a continuous block
+  // of numbers for array indexing in lookup tables. If changing State or
+  // Transition, don't forget to update k*Size and the lookup tables.
+  enum State {
+    kStart = 0,

[tkent, 2012/08/27 07:09:17]

See http://www.chromium.org/developers/coding-style
Also, It's hard to understand the difference between kCR1 and kCR. I prefer
adding prefixes to them.  e.g. STATE_CR TRANSITION_CR

[vabr (Chromium), 2012/08/29 19:57:07]

Added STATE_ prefix to states, transitions disappeared in the meantime.

+    kCR1 = 1,
+    kIgnorePreamble = 2,
+    kDB1 = 3,
+    kCR2 = 4,
+    kPart = 5,
+    kName = 6,
+    kColonS = 7,  // "S" to distinguish it from the transition kColonT.
+    kEnd1 = 8,
+    kEnd2 = 9,
+    kEnd3 = 10,
+    kCR3 = 11,
+    kPreData = 12,
+    kCR4 = 13,
+    kData = 14,
+    kData2 = 15,
+    kDB2 = 16,
+    kD = 17,
+    kEnd = 18,
+    kCR5 = 19,
+    kIgnoreEpilogue = 20,
+    kError = 21
+  };
+  enum Transition {
+    kLF = 0,
+    kCR = 1,
+    kAscii = 2,        // A "shorthand" for ASCII 33-126 without ':'.
+    kLwsp = 3,
+    kDashBoundary = 4,
+    kColonT = 5,       // "T" to distinguish it from the state kColonS.
+    kDash = 6,         // Meaning '-', not "--".
+    kAny = 7           // To represent '*'.
+  };
+  static const size_t kStateSize = 22;
+  static const size_t kTransitionSize = 8;
+
+  // Lookup tables:
+  // Maps transitions with one-character label to that character (else to 0).
+  static char kTransitionToChar[];
+  // Indices of transitions available in state |s| in |kAvailableTransitions|
+  // start at kStateToTransition[s] and the last transition for |s| is always
+  // kAny. The target state corresponding to transition kAvailableTransitions[i]
+  // is kNextState[i].
+  static Transition kAvailableTransitions[];
+  static State kNextState[];
+  static size_t kStateToTransition[];
+
+  // Reads the source until the next name-value pair is read. Returns true if
+  // |next_name_| and |next_value_| were successfully updated.
+  bool ReadNextNameValue();
+  // One step of the automaton, based on |state_| and the input from |source_|
+  // to be read. Updates the |offset_| iterator. Returns true on success.
+  bool DoStep();
+  // Tests whether the input pointed to by |offset_| allows to read transition
+  // |t|. It returns the number of bytes to be read, or 0 if |t| cannot be read.
+  size_t LookUp(Transition t);
+
+  // Extracts "name" and possibly "value" from a Content-Disposition header.
+  // Writes directly into |next_name_| and |next_value_|. Returns true on
+  // success and false otherwise.
+  bool ParseHeader(const base::StringPiece& header);
+
+  bool InFinalState() {
+    return state_ == kEnd || state_ == kIgnoreEpilogue;
+  }
+
+  // The parsed message can be split into multiple sources which we read
+  // sequentially.
+  base::StringPiece source_;
+  const char* source_end_;
+  const char* offset_;

[tkent, 2012/08/27 07:09:17]

The name "offset_" is confusing.  It's not an offset.  It's a cursor,
current-position, or something.

[vabr (Chromium), 2012/08/29 19:57:07]

You're right.
This disappeared after rewriting.

+  // The dash-boundary string is used for all sources.
+  const std::string dash_boundary_;
+  State state_;
+  // The next result to be returned by GetNextNameValue. It is stored as a pair
+  // of StringPieces instead of a Result, to avoid one copy of the data (note
+  // that Result stores a copy of the data in std::string, whereas StringPiece
+  // is just a pointer to the data in |source_|).
+  base::StringPiece next_name_;
+  base::StringPiece next_value_;
+  bool value_name_present_;
+
+  DISALLOW_COPY_AND_ASSIGN(FormDataParserMultipart);
+};
+
+// Implementation of FormDataParser and FormDataParser::Result .
+
+FormDataParser::Result::Result() {}
+FormDataParser::Result::~Result() {}
+
+void FormDataParser::Result::Reset() {
+  name_.erase();
+  value_.erase();
+}
+
+FormDataParser::~FormDataParser() {}
+
+// static
+scoped_ptr<FormDataParser> FormDataParser::Create(
+    const net::URLRequest* request) {
+  std::string value;
+  const bool found = request->extra_request_headers().GetHeader(
+      net::HttpRequestHeaders::kContentType, &value);
+  return Create(found ? &value : NULL);
+}
+
+// static
+scoped_ptr<FormDataParser> FormDataParser::Create(
+    const std::string* content_type_header) {
+  enum ParserChoice {kUrlEncoded, kMultipart, kError};
+  ParserChoice choice = kError;
+  std::string boundary;
+
+  if (content_type_header == NULL) {
+    choice = kUrlEncoded;
+  } else {
+    const std::string content_type(
+        content_type_header->substr(0, content_type_header->find(';')));
+
+    if (base::strcasecmp(
+        content_type.c_str(), "application/x-www-form-urlencoded") == 0) {
+      choice = kUrlEncoded;
+    } else if (base::strcasecmp(
+        content_type.c_str(), "multipart/form-data") == 0) {
+      static const char kBoundaryString[] = "boundary=";
+      size_t offset = content_type_header->find(kBoundaryString);
+      if (offset == std::string::npos) {
+        // Malformed header.
+        return scoped_ptr<FormDataParser>();
+      }
+      offset += strlen(kBoundaryString);
+      boundary = content_type_header->substr(
+          offset, content_type_header->find(';', offset));
+      if (!boundary.empty())
+        choice = kMultipart;
+    }
+  }
+  // Other cases are unparseable, including when |content_type| is "text/plain".
+
+  switch (choice) {
+    case kUrlEncoded:
+      return scoped_ptr<FormDataParser>(new FormDataParserUrlEncoded());
+    case kMultipart:
+      return scoped_ptr<FormDataParser>(new FormDataParserMultipart(boundary));
+    default:  // In other words, case kError:
+      return scoped_ptr<FormDataParser>();
+  }
+}
+
+FormDataParser::FormDataParser() {}
+
+// Implementation of FormDataParserUrlEncoded.
+
+FormDataParserUrlEncoded::FormDataParserUrlEncoded()
+    : source_end_(NULL),
+      aborted_(false),
+      offset_(NULL),
+      equality_signs_(0),
+      amp_signs_(0),
+      expect_equality_(true) {
+}
+
+FormDataParserUrlEncoded::~FormDataParserUrlEncoded() {}
+
+bool FormDataParserUrlEncoded::AllDataReadOK() {
+  return source_.data() != NULL &&
+         !aborted_ &&
+         offset_ == source_end_ &&
+         equality_signs_ == amp_signs_ + 1;

[tkent, 2012/08/27 07:09:17]

Why do we need to check the number of = and & ?
equality_signs_ and amp_signs_ make GetNextChar complex.

[vabr (Chromium), 2012/08/29 19:57:07]

Now the parser uses a regexp which eliminates such bookkeeping.

+}
+
+bool FormDataParserUrlEncoded::GetNextNameValue(Result* result) {
+  result->Reset();
+  if (source_.data() == NULL || aborted_)
+    return false;
+  if (offset_ == source_end_)
+    return false;
+  const char* const name_start = &(*offset_);
+  char c;
+  bool last_read_success = GetNextChar(&c);
+  while (last_read_success && c != '=')
+    last_read_success = GetNextChar(&c);
+  if (!last_read_success) {  // This means the data is malformed.
+    Abort();
+    return false;
+  }
+  const char* const name_end = offset_ - 1;
+  const std::string encoded_name(name_start, name_end - name_start);
+  const net::UnescapeRule::Type unescape_rules =
+      net::UnescapeRule::URL_SPECIAL_CHARS | net::UnescapeRule::CONTROL_CHARS |
+      net::UnescapeRule::SPACES | net::UnescapeRule::REPLACE_PLUS_WITH_SPACE;
+  result->set_name(net::UnescapeURLComponent(encoded_name, unescape_rules));
+
+  const char* const value_start = offset_;
+  last_read_success = GetNextChar(&c);
+  while (last_read_success && c != '&')
+    last_read_success = GetNextChar(&c);
+  const char* const value_end =
+      last_read_success ? offset_ - 1 : offset_;
+  const std::string encoded_value(value_start, value_end - value_start);
+  result->set_value(net::UnescapeURLComponent(encoded_value, unescape_rules));
+  return true;
+}
+
+bool FormDataParserUrlEncoded::SetSource(const base::StringPiece& source) {
+  if (source_ != NULL || source.data() == NULL || aborted_)
+    return false;
+  source_ = source;
+  source_end_ = source_.data() + source_.size();
+  offset_ = source_.data();
+  return true;
+}
+
+bool FormDataParserUrlEncoded::GetNextChar(char* c) {
+  if (offset_ == source_end_ || aborted_)
+    return false;
+  *c = *offset_;
+  ++offset_;
+
+  if (*c == '=') {
+    if (expect_equality_) {
+      ++equality_signs_;
+      expect_equality_ = false;
+    } else {
+      Abort();
+      return false;
+    }
+  }
+  if (*c == '&' && offset_ != source_end_) {
+    if (!expect_equality_) {
+      ++amp_signs_;
+      expect_equality_ = true;
+    } else {
+      Abort();
+      return false;
+    }
+  }
+
+  return true;
+}
+
+void FormDataParserUrlEncoded::Abort() {
+  aborted_ = true;
+}
+
+// Implementation of FormDataParserMultipart.
+
+FormDataParserMultipart::FormDataParserMultipart(
+    const std::string& boundary_separator)
+    : source_end_(NULL),
+      offset_(NULL),
+      dash_boundary_("--" + boundary_separator),
+      state_(kStart),
+      value_name_present_(false) {
+}
+
+FormDataParserMultipart::~FormDataParserMultipart() {}
+
+bool FormDataParserMultipart::AllDataReadOK() {
+  return source_.data() != NULL && InFinalState();
+}
+
+bool FormDataParserMultipart::GetNextNameValue(Result* result) {
+  if (!value_name_present_ || state_ == kError)
+    return false;
+  result->set_name(next_name_);
+  result->set_value(next_value_);
+  next_name_.clear();
+  next_value_.clear();
+  value_name_present_ = ReadNextNameValue();
+  return true;
+}
+
+bool FormDataParserMultipart::SetSource(const base::StringPiece& source) {
+  if (state_ == kError ||
+      source.data() == NULL ||
+      // Message part across a source split is also an error.
+      next_name_.data() != NULL || next_value_.data() != NULL)
+    return false;
+  if (source_.data() != NULL && offset_ != source_end_){
+    // Try to seek until the end. If no name-value pair is found, this is OK.
+    value_name_present_ = ReadNextNameValue();
+    if (!value_name_present_ || offset_ != source_end_)
+      return false;
+  }
+  source_ = source;
+  source_end_ = source_.data() + source_.size();
+  offset_ = source_.data();
+  value_name_present_ = ReadNextNameValue();
+  return true;
+}
+
+// static
+char FormDataParserMultipart::kTransitionToChar[] = {
+  '\n',   // For kLF.
+  '\r',   // For kCR.
+  0,      // For kAscii.
+  0,      // For kLwsp.
+  0,      // For kDashBoundary.
+  ':',    // For kColonT.
+  '-',    // For kDash.
+  0,      // For kAny.
+};
+
+// static
+FormDataParserMultipart::Transition
+    FormDataParserMultipart::kAvailableTransitions[] = {
+  kDashBoundary, kCR, kAny,     // For kStart.
+  kLF, kAny,                    // For kCR1.
+  kCR, kAny,                    // For kIgnorePreamble.
+  kLwsp, kCR, kAny,             // For kDB1.
+  kLF, kAny,                    // For kCR2.
+  kAscii, kCR, kAny,            // For kPart.
+  kAscii, kColonT, kAny,        // For kName.
+  kLF, kCR, kAny,               // For kColonS.
+  kCR, kAscii, kAny,            // For kEnd1.
+  kLF, kCR, kAscii, kAny,       // For kEnd2.
+  kLwsp, kCR, kAscii, kAny,     // For kEnd3.
+  kLF, kAny,                    // For kCR3.
+  kDashBoundary, kCR, kAny,     // For kPreData.
+  kLF, kAny,                    // For kCR4.
+  kCR, kAny,                    // For kData.
+  kDashBoundary, kAny,          // For kData2.
+  kLwsp, kCR, kDash, kAny,      // For kDB2.
+  kDash, kAny,                  // For kD.
+  kLwsp, kCR, kAny,             // For kEnd.
+  kLF, kAny,                    // For kCR5.
+  kAny,                         // For kIgnoreEpilogue.
+  kAny                          // For kError.
+};
+
+// static
+FormDataParserMultipart::State FormDataParserMultipart::kNextState[] = {

[tkent, 2012/08/27 07:09:17]

kAvailableTransitions and kNextState should be merged into one.
like:

  {kDashBoundary, kDB1}, {kCR, kCR1}, {kAny, kIgnorePreamble},
  ....

[vabr (Chromium), 2012/08/29 19:57:07]

Disappeared after rewriting.

+  kDB1, kCR1, kIgnorePreamble,   // For kStart;         size so far:    03.
+  kStart, kIgnorePreamble,       // For kCR1;                           05.
+  kCR1, kIgnorePreamble,         // For kIgnorePreamble;                07.
+  kDB1, kCR2, kError,            // For kDB1;                           10.
+  kPart, kError,                 // For kCR2;                           12.
+  kName, kCR3, kError,           // For kPart;                          15.
+  kName, kColonS, kError,        // For kName;                          18.
+  kEnd1, kEnd2, kColonS,         // For kColonS;                        21.
+  kCR3, kName, kError,           // For kEnd1;                          24.
+  kEnd3, kCR3, kName, kError,    // For kEnd2;                          28.
+  kColonS, kCR3, kName, kError,  // For kEnd3;                          32.
+  kPreData, kError,              // For kCR3;                           34.
+  kDB2, kCR3, kData,             // For kPreData;                       37.
+  kData2, kData,                 // For kCR4;                           39.
+  kCR4, kData,                   // For kData;                          41.
+  kDB2, kCR4,                    // For kData2;                         43.
+  kDB1, kCR2, kD, kError,        // For kDB2;                           47.
+  kEnd, kError,                  // For kD;                             49.
+  kEnd, kCR5, kError,            // For kEnd;                           52.
+  kIgnoreEpilogue, kError,       // For kCR5;                           54.
+  kIgnoreEpilogue,               // For kIgnoreEpilogue;                55.
+  kError                         // For kError;                         56.
+};
+
+// static
+size_t FormDataParserMultipart::kStateToTransition[] = {
+  0u,  // For kStart
+  3u,  // For kCR1
+  5u,  // For kIgnorePreamble
+  7u,  // For kDB1
+  10u, // For kCR2
+  12u, // For kPart
+  15u, // For kName
+  18u, // For kColonS
+  21u, // For kEnd1
+  24u, // For kEnd2
+  28u, // For kEnd3
+  32u, // For kCR3
+  34u, // For kPreData
+  37u, // For kCR4
+  39u, // For kData
+  41u, // For kData2
+  43u, // For kDB2
+  47u, // For kD
+  49u, // For kEnd
+  52u, // For kCR5
+  54u, // For kIgnoreEpilogue
+  55u, // For kError
+};
+
+bool FormDataParserMultipart::ReadNextNameValue() {
+  if (state_ == kError || source_.data() == NULL ||
+      next_name_.data() != NULL || next_value_.data() != NULL)
+    return false;
+
+  // Seek to the next part's headers.
+  while (state_ != kPart) {
+    if (!DoStep())
+      return false;
+  }

[tkent, 2012/08/27 07:09:17]

We had better have a function DoStepsUntil(State). It will improve code
readability.

if (!DoStepsUntil(kPart))
    return false;

[vabr (Chromium), 2012/08/29 19:57:07]

Disappeared after the rewrite.

+  while (state_ != kPreData) {
+    const char* header = offset_;
+    while (state_ != kColonS) {
+      if (!DoStep())
+        return false;
+    }
+    size_t header_length = 0u;
+    while (state_ != kPreData && state_ != kName) {
+      if (state_ == kEnd1 || state_ == kEnd2 || state_ == kEnd3) {
+        // The cast is safe, we know that offset only moves forward.
+        header_length = static_cast<size_t>(offset_ - header);
+      }
+      if (!DoStep())
+        return false;
+    }
+    if (ParseHeader(base::StringPiece(header, header_length))) {
+      // Found what we were looking for, just skip to the part's body.
+      while (state_ != kPreData) {
+        if (!DoStep())
+          return false;
+      }
+    }
+  }
+
+  const char* body = offset_;
+  size_t body_length = 0;
+  while (state_ != kDB2 && offset_ != source_end_) {
+    if (!DoStep())
+      return false;
+    if (state_ == kCR4) {
+      // We are in the middle of which might be the CRLF starting the part
+      // separator (see the "delimiter" non-terminal from the grammar given
+      // in the header file). The cast is safe, we know that offset only moves
+      // forward and body was assigned at least 1 transition ago.
+      body_length = static_cast<size_t>(offset_ - body - 1);
+    }
+  }
+  if (body_length > 0)
+    next_value_.set(body, body_length);
+  return true;
+}
+
+bool FormDataParserMultipart::DoStep() {
+  if (state_ == kError || offset_ == source_end_)
+    return false;
+  size_t transition_index = kStateToTransition[state_];
+  Transition t = kAvailableTransitions[transition_index];
+  while (t != kAny) {
+    const State s = kNextState[transition_index];
+    const size_t length = LookUp(t);
+    if (length > 0) {
+      offset_ += length;
+      state_ = s;
+      return true;
+    }
+    t = kAvailableTransitions[++transition_index];
+  }
+  // We have kAny, the default choice. Seek by one and switch the state.
+  ++offset_;
+  state_ = kNextState[transition_index];
+  return true;
+}
+
+// Contract -- the following must be true: offset_ != source_end_ .
+// The idea is to check this only once in the caller (DoStep()), and do not
+// repeat it here every time, as this can be called many times from one call
+// to DoStep().
+size_t FormDataParserMultipart::LookUp(FormDataParserMultipart::Transition t) {
+  const char ahead = *offset_;
+  const char first_char = kTransitionToChar[t];
+
+  // Easy case: labels corresponding to a single char.
+  if (first_char != 0)
+    return ahead == first_char ? 1u : 0u;
+
+  // Harder cases.
+  switch (t) {
+    // Multiple alternatives, 1-char long: return immediately.
+    case kAscii:
+      return (ahead >= 33 && ahead <= 126 && ahead != ':') ? 1u : 0u;
+    case kLwsp:
+      return (ahead == ' ' || ahead == '\t') ? 1u : 0u;
+
+    // Longer than 1 char: prepare work for later.
+    case kDashBoundary: {
+      const size_t length = dash_boundary_.size();
+      // The cast below is safe, we know that the difference is not negative.
+      if (static_cast<size_t>(source_end_ - offset_) < length ||
+          memcmp(dash_boundary_.c_str(), offset_, length) != 0)
+        return 0u;
+      return length;
+    }
+    case kAny:
+      // We are not supposed to be asked for kAny, but this is the right answer:
+      return 1u;
+    default:  // We never get here -- the rest has already been handled above.
+      NOTREACHED();
+      return 0u;
+  }
+}
+
+bool FormDataParserMultipart::ParseHeader(const base::StringPiece& header) {
+  static const char kContentDisposition[] = "Content-Disposition:";
+  if (memcmp(header.data(), kContentDisposition,
+             strlen(kContentDisposition) != 0))

[tkent, 2012/08/27 07:09:17]

strlen is not needed.  The length of kContentDisposition is fixed.

[vabr (Chromium), 2012/08/29 19:57:07]

Correct. Although this particular instance and those pointed below disappeared,
I will re-check the whole CL, as there are many instances of this.

+    return false;
+  static const char kNameEquals[] = " name=\"";
+  static const char kFilenameEquals[] = " filename=\"";
+
+  // Mandatory part: find the name and set it as |next_name_|.
+  StringPiece::size_type field_offset = header.find(kNameEquals);
+  if (field_offset == StringPiece::npos)
+    return false;
+  field_offset += strlen(kNameEquals);

[tkent, 2012/08/27 07:09:17]

ditto.

+  StringPiece::size_type field_end = header.find('"', field_offset);
+  if (field_end == StringPiece::npos)
+    return false;
+  next_name_.set(header.data() + field_offset, field_end - field_offset);

[tkent, 2012/08/27 07:09:17]

Need to decode the name value.

BTW, what's the expected character encoding of next_name_ ?
Will it be passed to an extension as JavaScript string?

I remember we have no ways to detect character encoding of submitted form data.
If a page containing the form is encoded in ISO-8859-1, names and values are
encoded in ISO-8859-1, and if a page is encoded in UTF-8, names and values are
encoded in UTF-8.

[vabr (Chromium), 2012/08/29 19:57:07]

Thanks very much for bringing this up!

On 2012/08/27 07:09:17, Kent Tamura wrote:
I am not sure what you mean by "data".
The text present in the HTML file itself, i.e., the field names and default
values, are, as far as I saw, uploaded without any change. If the HTML file is
written in UTF8, the names and default values are uploaded in UTF8, even if the
HTML page itself specifies a different content-type encoding via its meta tag
"http-equiv".
The values entered into the form by the user, on the other hand, seem to follow
the declared content-type encoding of the page, not the encoding of the HTML
file.
Yes, it will be passed to the extensions byte-per-byte as a JavaScript string
(StringValue).
If both the page encoding and the HTML file are UTF8 then there is no need to
decode the name value, it will safely reach the extension. If it is not in UTF8
then parsing will fail, because re2 assumes UTF8 data, and also StringValue
checks on construction whether the string is a valid UTF8 string
(IsStringUTF8()). In that case the form will not be parsed and the extension
gets the unparsed upload data via a different part of code from this CL.

If there is no way to detect the character encoding, then I don't see how I
could support the non-UTF8 cases. Speaking with battre@, we consider it
acceptable to have form parsing only for UTF8 websites. Based on that, I suggest
doing no decoding here. What do you think?

+
+  // Optional part: find the filename and set it as |next_value_|.
+  field_offset = header.find(kFilenameEquals);
+  if (field_offset == StringPiece::npos)
+    return true;  // This was only optional
+  field_offset += strlen(kFilenameEquals);

[tkent, 2012/08/27 07:09:17]

ditto.

+  field_end = header.find('"', field_offset);
+  if (field_end == StringPiece::npos)
+    return false;  // This is a malformed header.
+  next_value_.set(header.data() + field_offset, field_end - field_offset);
+  return true;
+}
+
+}  // namespace extensions