blob: 9cffa34ddb880babd099110af6e4de69fe161ca5 [file] [log] [blame]
// Copyright 2016 PDFium Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
// Original code copyright 2014 Foxit Software Inc. http://www.foxitsoftware.com
#include "core/fpdftext/cpdf_linkextract.h"
#include <vector>
#include "core/fpdftext/cpdf_textpage.h"
#include "core/fxcrt/fx_extension.h"
#include "core/fxcrt/fx_string.h"
#include "core/fxcrt/fx_system.h"
namespace {
// Find the end of a web link starting from offset |start| and ending at offset
// |end|. The purpose of this function is to separate url from the surrounding
// context characters, we do not intend to fully validate the url. |str|
// contains lower case characters only.
size_t FindWebLinkEnding(const WideString& str, size_t start, size_t end) {
if (str.Contains(L'/', start)) {
// When there is a path and query after '/', most ASCII chars are allowed.
// We don't sanitize in this case.
return end;
}
// When there is no path, it only has IP address or host name.
// Port is optional at the end.
if (str[start] == L'[') {
// IPv6 reference.
// Find the end of the reference.
auto result = str.Find(L']', start + 1);
if (result.has_value()) {
end = result.value();
if (end > start + 1) { // Has content inside brackets.
size_t len = str.GetLength();
size_t off = end + 1;
if (off < len && str[off] == L':') {
off++;
while (off < len && FXSYS_IsDecimalDigit(str[off]))
off++;
if (off > end + 2 &&
off <= len) // At least one digit in port number.
end = off - 1; // |off| is offset of the first invalid char.
}
}
}
return end;
}
// According to RFC1123, host name only has alphanumeric chars, hyphens,
// and periods. Hyphen should not at the end though.
// Non-ASCII chars are ignored during checking.
while (end > start && str[end] < 0x80) {
if (FXSYS_IsDecimalDigit(str[end]) ||
(str[end] >= L'a' && str[end] <= L'z') || str[end] == L'.') {
break;
}
end--;
}
return end;
}
// Remove characters from the end of |str|, delimited by |start| and |end|, up
// to and including |charToFind|. No-op if |charToFind| is not present. Updates
// |end| if characters were removed.
void TrimBackwardsToChar(const WideString& str,
wchar_t charToFind,
size_t start,
size_t* end) {
for (size_t pos = *end; pos >= start; pos--) {
if (str[pos] == charToFind) {
*end = pos - 1;
break;
}
}
}
// Finds opening brackets ()[]{}<> and quotes "' before the URL delimited by
// |start| and |end| in |str|. Matches a closing bracket or quote for each
// opening character and, if present, removes everything afterwards. Returns the
// new end position for the string.
size_t TrimExternalBracketsFromWebLink(const WideString& str,
size_t start,
size_t end) {
for (size_t pos = 0; pos < start; pos++) {
if (str[pos] == '(') {
TrimBackwardsToChar(str, ')', start, &end);
} else if (str[pos] == '[') {
TrimBackwardsToChar(str, ']', start, &end);
} else if (str[pos] == '{') {
TrimBackwardsToChar(str, '}', start, &end);
} else if (str[pos] == '<') {
TrimBackwardsToChar(str, '>', start, &end);
} else if (str[pos] == '"') {
TrimBackwardsToChar(str, '"', start, &end);
} else if (str[pos] == '\'') {
TrimBackwardsToChar(str, '\'', start, &end);
}
}
return end;
}
} // namespace
CPDF_LinkExtract::CPDF_LinkExtract(const CPDF_TextPage* pTextPage)
: m_pTextPage(pTextPage) {}
CPDF_LinkExtract::~CPDF_LinkExtract() = default;
void CPDF_LinkExtract::ExtractLinks() {
m_LinkArray.clear();
size_t start = 0;
size_t pos = 0;
bool bAfterHyphen = false;
bool bLineBreak = false;
const size_t nTotalChar = m_pTextPage->CountChars();
const WideString page_text = m_pTextPage->GetAllPageText();
while (pos < nTotalChar) {
const CPDF_TextPage::CharInfo& char_info = m_pTextPage->GetCharInfo(pos);
if (char_info.m_CharType != CPDF_TextPage::CharType::kGenerated &&
char_info.m_Unicode != L' ' && pos != nTotalChar - 1) {
bAfterHyphen =
(char_info.m_CharType == CPDF_TextPage::CharType::kHyphen ||
(char_info.m_CharType == CPDF_TextPage::CharType::kNormal &&
char_info.m_Unicode == L'-'));
++pos;
continue;
}
size_t nCount = pos - start;
if (pos == nTotalChar - 1) {
++nCount;
} else if (bAfterHyphen &&
(char_info.m_Unicode == L'\n' || char_info.m_Unicode == L'\r')) {
// Handle text breaks with a hyphen to the next line.
bLineBreak = true;
++pos;
continue;
}
WideString strBeCheck = page_text.Substr(start, nCount);
if (bLineBreak) {
strBeCheck.Remove(L'\n');
strBeCheck.Remove(L'\r');
bLineBreak = false;
}
// Replace the generated code with the hyphen char.
strBeCheck.Replace(L"\xfffe", L"-");
if (strBeCheck.GetLength() > 5) {
while (strBeCheck.GetLength() > 0) {
wchar_t ch = strBeCheck.Back();
if (ch != L')' && ch != L',' && ch != L'>' && ch != L'.')
break;
strBeCheck = strBeCheck.First(strBeCheck.GetLength() - 1);
nCount--;
}
// Check for potential web URLs and email addresses.
// Ftp address, file system links, data, blob etc. are not checked.
if (nCount > 5) {
auto maybe_link = CheckWebLink(strBeCheck);
if (maybe_link.has_value()) {
maybe_link.value().m_Start += start;
m_LinkArray.push_back(maybe_link.value());
} else if (CheckMailLink(&strBeCheck)) {
m_LinkArray.push_back(Link{{start, nCount}, strBeCheck});
}
}
}
start = ++pos;
}
}
absl::optional<CPDF_LinkExtract::Link> CPDF_LinkExtract::CheckWebLink(
const WideString& strBeCheck) {
static const wchar_t kHttpScheme[] = L"http";
static const wchar_t kWWWAddrStart[] = L"www.";
const size_t kHttpSchemeLen = FXSYS_len(kHttpScheme);
const size_t kWWWAddrStartLen = FXSYS_len(kWWWAddrStart);
WideString str = strBeCheck;
str.MakeLower();
// First, try to find the scheme.
auto start = str.Find(kHttpScheme);
if (start.has_value()) {
size_t off = start.value() + kHttpSchemeLen; // move after "http".
if (str.GetLength() > off + 4) { // At least "://<char>" follows.
if (str[off] == L's') // "https" scheme is accepted.
off++;
if (str[off] == L':' && str[off + 1] == L'/' && str[off + 2] == L'/') {
off += 3;
const size_t end =
FindWebLinkEnding(str, off,
TrimExternalBracketsFromWebLink(
str, start.value(), str.GetLength() - 1));
if (end > off) { // Non-empty host name.
const size_t nStart = start.value();
const size_t nCount = end - nStart + 1;
return Link{{nStart, nCount}, strBeCheck.Substr(nStart, nCount)};
}
}
}
}
// When there is no scheme, try to find url starting with "www.".
start = str.Find(kWWWAddrStart);
if (start.has_value()) {
size_t off = start.value() + kWWWAddrStartLen;
if (str.GetLength() > off) {
const size_t end =
FindWebLinkEnding(str, start.value(),
TrimExternalBracketsFromWebLink(
str, start.value(), str.GetLength() - 1));
if (end > off) {
const size_t nStart = start.value();
const size_t nCount = end - nStart + 1;
return Link{{nStart, nCount},
L"http://" + strBeCheck.Substr(nStart, nCount)};
}
}
}
return absl::nullopt;
}
bool CPDF_LinkExtract::CheckMailLink(WideString* str) {
auto aPos = str->Find(L'@');
// Invalid when no '@' or when starts/ends with '@'.
if (!aPos.has_value() || aPos.value() == 0 || aPos == str->GetLength() - 1)
return false;
// Check the local part.
size_t pPos = aPos.value(); // Used to track the position of '@' or '.'.
for (size_t i = aPos.value(); i > 0; i--) {
wchar_t ch = (*str)[i - 1];
if (ch == L'_' || ch == L'-' || FXSYS_iswalnum(ch))
continue;
if (ch != L'.' || i == pPos || i == 1) {
if (i == aPos.value()) {
// There is '.' or invalid char before '@'.
return false;
}
// End extracting for other invalid chars, '.' at the beginning, or
// consecutive '.'.
size_t removed_len = i == pPos ? i + 1 : i;
*str = str->Last(str->GetLength() - removed_len);
break;
}
// Found a valid '.'.
pPos = i - 1;
}
// Check the domain name part.
aPos = str->Find(L'@');
if (!aPos.has_value() || aPos.value() == 0)
return false;
str->TrimRight(L'.');
// At least one '.' in domain name, but not at the beginning.
// TODO(weili): RFC5322 allows domain names to be a local name without '.'.
// Check whether we should remove this check.
auto ePos = str->Find(L'.', aPos.value() + 1);
if (!ePos.has_value() || ePos.value() == aPos.value() + 1)
return false;
// Validate all other chars in domain name.
size_t nLen = str->GetLength();
pPos = 0; // Used to track the position of '.'.
for (size_t i = aPos.value() + 1; i < nLen; i++) {
wchar_t wch = (*str)[i];
if (wch == L'-' || FXSYS_iswalnum(wch))
continue;
if (wch != L'.' || i == pPos + 1) {
// Domain name should end before invalid char.
size_t host_end = i == pPos + 1 ? i - 2 : i - 1;
if (pPos > 0 && host_end - aPos.value() >= 3) {
// Trim the ending invalid chars if there is at least one '.' and name.
*str = str->First(host_end + 1);
break;
}
return false;
}
pPos = i;
}
if (!str->Contains(L"mailto:"))
*str = L"mailto:" + *str;
return true;
}
WideString CPDF_LinkExtract::GetURL(size_t index) const {
return index < m_LinkArray.size() ? m_LinkArray[index].m_strUrl
: WideString();
}
std::vector<CFX_FloatRect> CPDF_LinkExtract::GetRects(size_t index) const {
if (index >= m_LinkArray.size())
return std::vector<CFX_FloatRect>();
return m_pTextPage->GetRectArray(m_LinkArray[index].m_Start,
m_LinkArray[index].m_Count);
}
absl::optional<CPDF_LinkExtract::Range> CPDF_LinkExtract::GetTextRange(
size_t index) const {
if (index >= m_LinkArray.size())
return absl::nullopt;
return m_LinkArray[index];
}