components/visitedlink/common/visitedlink_common.cc

// Copyright (c) 2011 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 "components/visitedlink/common/visitedlink_common.h"

#include <string.h>  // for memset()

#include "base/bit_cast.h"
#include "base/hash/md5.h"
#include "base/logging.h"
#include "url/gurl.h"

namespace visitedlink {

const VisitedLinkCommon::Fingerprint VisitedLinkCommon::null_fingerprint_ = 0;
const VisitedLinkCommon::Hash VisitedLinkCommon::null_hash_ = -1;

VisitedLinkCommon::VisitedLinkCommon()
    : hash_table_(nullptr), table_length_(0) {
  memset(salt_, 0, sizeof(salt_));
}

VisitedLinkCommon::~VisitedLinkCommon() {
}

// FIXME: this uses linear probing, it should be replaced with quadratic
// probing or something better. See VisitedLinkMaster::AddFingerprint
bool VisitedLinkCommon::IsVisited(const char* canonical_url,
                                  size_t url_len) const {
  if (url_len == 0)
    return false;
  if (!hash_table_ || table_length_ == 0)
    return false;
  return IsVisited(ComputeURLFingerprint(canonical_url, url_len));
}

bool VisitedLinkCommon::IsVisited(const GURL& url) const {
  return IsVisited(url.spec().data(), url.spec().size());
}

bool VisitedLinkCommon::IsVisited(Fingerprint fingerprint) const {
  // Go through the table until we find the item or an empty spot (meaning it
  // wasn't found). This loop will terminate as long as the table isn't full,
  // which should be enforced by AddFingerprint.
  Hash first_hash = HashFingerprint(fingerprint);
  Hash cur_hash = first_hash;
  while (true) {
    Fingerprint cur_fingerprint = FingerprintAt(cur_hash);
    if (cur_fingerprint == null_fingerprint_)
      return false;  // End of probe sequence found.
    if (cur_fingerprint == fingerprint)
      return true;  // Found a match.

    // This spot was taken, but not by the item we're looking for, search in
    // the next position.
    cur_hash++;
    if (cur_hash == table_length_)
      cur_hash = 0;
    if (cur_hash == first_hash) {
      // Wrapped around and didn't find an empty space, this means we're in an
      // infinite loop because AddFingerprint didn't do its job resizing.
      NOTREACHED();
      return false;
    }
  }
}

// Uses the top 64 bits of the MD5 sum of the canonical URL as the fingerprint,
// this is as random as any other subset of the MD5SUM.
//
// FIXME: this uses the MD5SUM of the 16-bit character version. For systems
// where wchar_t is not 16 bits (Linux uses 32 bits, I think), this will not be
// compatable. We should define explicitly what should happen here across
// platforms, and convert if necessary (probably to UTF-16).

// static
VisitedLinkCommon::Fingerprint VisitedLinkCommon::ComputeURLFingerprint(
    const char* canonical_url,
    size_t url_len,
    const uint8_t salt[LINK_SALT_LENGTH]) {
  DCHECK(url_len > 0) << "Canonical URLs should not be empty";

  base::MD5Context ctx;
  base::MD5Init(&ctx);
  base::MD5Update(&ctx, base::StringPiece(reinterpret_cast<const char*>(salt),
                                          LINK_SALT_LENGTH));
  base::MD5Update(&ctx, base::StringPiece(canonical_url, url_len));

  base::MD5Digest digest;
  base::MD5Final(&digest, &ctx);

  // This is the same as "return *(Fingerprint*)&digest.a;" but if we do that
  // direct cast the alignment could be wrong, and we can't access a 64-bit int
  // on arbitrary alignment on some processors. This reinterpret_casts it
  // down to a char array of the same size as fingerprint, and then does the
  // bit cast, which amounts to a memcpy. This does not handle endian issues.
  return bit_cast<Fingerprint, uint8_t[8]>(
      *reinterpret_cast<uint8_t(*)[8]>(&digest.a));
}

}  // namespace visitedlink