gecko-dev/toolkit/components/url-classifier/ProtocolParser.cpp

1090 строки
31 KiB
C++

//* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
/* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
#include "ProtocolParser.h"
#include "LookupCache.h"
#include "nsNetCID.h"
#include "mozilla/Logging.h"
#include "prnetdb.h"
#include "prprf.h"
#include "nsUrlClassifierUtils.h"
#include "nsPrintfCString.h"
#include "mozilla/Base64.h"
#include "RiceDeltaDecoder.h"
#include "mozilla/EndianUtils.h"
// MOZ_LOG=UrlClassifierProtocolParser:5
mozilla::LazyLogModule gUrlClassifierProtocolParserLog("UrlClassifierProtocolParser");
#define PARSER_LOG(args) MOZ_LOG(gUrlClassifierProtocolParserLog, mozilla::LogLevel::Debug, args)
namespace mozilla {
namespace safebrowsing {
// Updates will fail if fed chunks larger than this
const uint32_t MAX_CHUNK_SIZE = (1024 * 1024);
// Updates will fail if the total number of tocuhed chunks is larger than this
const uint32_t MAX_CHUNK_RANGE = 1000000;
const uint32_t DOMAIN_SIZE = 4;
// Parse one stringified range of chunks of the form "n" or "n-m" from a
// comma-separated list of chunks. Upon return, 'begin' will point to the
// next range of chunks in the list of chunks.
static bool
ParseChunkRange(nsACString::const_iterator& aBegin,
const nsACString::const_iterator& aEnd,
uint32_t* aFirst, uint32_t* aLast)
{
nsACString::const_iterator iter = aBegin;
FindCharInReadable(',', iter, aEnd);
nsAutoCString element(Substring(aBegin, iter));
aBegin = iter;
if (aBegin != aEnd)
aBegin++;
uint32_t numRead = PR_sscanf(element.get(), "%u-%u", aFirst, aLast);
if (numRead == 2) {
if (*aFirst > *aLast) {
uint32_t tmp = *aFirst;
*aFirst = *aLast;
*aLast = tmp;
}
return true;
}
if (numRead == 1) {
*aLast = *aFirst;
return true;
}
return false;
}
///////////////////////////////////////////////////////////////
// ProtocolParser implementation
ProtocolParser::ProtocolParser()
: mUpdateStatus(NS_OK)
{
}
ProtocolParser::~ProtocolParser()
{
CleanupUpdates();
}
nsresult
ProtocolParser::Init(nsICryptoHash* aHasher)
{
mCryptoHash = aHasher;
return NS_OK;
}
void
ProtocolParser::CleanupUpdates()
{
for (uint32_t i = 0; i < mTableUpdates.Length(); i++) {
delete mTableUpdates[i];
}
mTableUpdates.Clear();
}
TableUpdate *
ProtocolParser::GetTableUpdate(const nsACString& aTable)
{
for (uint32_t i = 0; i < mTableUpdates.Length(); i++) {
if (aTable.Equals(mTableUpdates[i]->TableName())) {
return mTableUpdates[i];
}
}
// We free automatically on destruction, ownership of these
// updates can be transferred to DBServiceWorker, which passes
// them back to Classifier when doing the updates, and that
// will free them.
TableUpdate *update = CreateTableUpdate(aTable);
mTableUpdates.AppendElement(update);
return update;
}
///////////////////////////////////////////////////////////////////////
// ProtocolParserV2
ProtocolParserV2::ProtocolParserV2()
: mState(PROTOCOL_STATE_CONTROL)
, mUpdateWait(0)
, mResetRequested(false)
, mTableUpdate(nullptr)
{
}
ProtocolParserV2::~ProtocolParserV2()
{
}
void
ProtocolParserV2::SetCurrentTable(const nsACString& aTable)
{
auto update = GetTableUpdate(aTable);
mTableUpdate = TableUpdate::Cast<TableUpdateV2>(update);
}
nsresult
ProtocolParserV2::AppendStream(const nsACString& aData)
{
if (NS_FAILED(mUpdateStatus))
return mUpdateStatus;
nsresult rv;
mPending.Append(aData);
bool done = false;
while (!done) {
if (mState == PROTOCOL_STATE_CONTROL) {
rv = ProcessControl(&done);
} else if (mState == PROTOCOL_STATE_CHUNK) {
rv = ProcessChunk(&done);
} else {
NS_ERROR("Unexpected protocol state");
rv = NS_ERROR_FAILURE;
}
if (NS_FAILED(rv)) {
mUpdateStatus = rv;
return rv;
}
}
return NS_OK;
}
void
ProtocolParserV2::End()
{
// Inbound data has already been processed in every AppendStream() call.
}
nsresult
ProtocolParserV2::ProcessControl(bool* aDone)
{
nsresult rv;
nsAutoCString line;
*aDone = true;
while (NextLine(line)) {
PARSER_LOG(("Processing %s\n", line.get()));
if (StringBeginsWith(line, NS_LITERAL_CSTRING("i:"))) {
// Set the table name from the table header line.
SetCurrentTable(Substring(line, 2));
} else if (StringBeginsWith(line, NS_LITERAL_CSTRING("n:"))) {
if (PR_sscanf(line.get(), "n:%d", &mUpdateWait) != 1) {
PARSER_LOG(("Error parsing n: '%s' (%d)", line.get(), mUpdateWait));
return NS_ERROR_FAILURE;
}
} else if (line.EqualsLiteral("r:pleasereset")) {
mResetRequested = true;
} else if (StringBeginsWith(line, NS_LITERAL_CSTRING("u:"))) {
rv = ProcessForward(line);
NS_ENSURE_SUCCESS(rv, rv);
} else if (StringBeginsWith(line, NS_LITERAL_CSTRING("a:")) ||
StringBeginsWith(line, NS_LITERAL_CSTRING("s:"))) {
rv = ProcessChunkControl(line);
NS_ENSURE_SUCCESS(rv, rv);
*aDone = false;
return NS_OK;
} else if (StringBeginsWith(line, NS_LITERAL_CSTRING("ad:")) ||
StringBeginsWith(line, NS_LITERAL_CSTRING("sd:"))) {
rv = ProcessExpirations(line);
NS_ENSURE_SUCCESS(rv, rv);
}
}
*aDone = true;
return NS_OK;
}
nsresult
ProtocolParserV2::ProcessExpirations(const nsCString& aLine)
{
if (!mTableUpdate) {
NS_WARNING("Got an expiration without a table.");
return NS_ERROR_FAILURE;
}
const nsCSubstring &list = Substring(aLine, 3);
nsACString::const_iterator begin, end;
list.BeginReading(begin);
list.EndReading(end);
while (begin != end) {
uint32_t first, last;
if (ParseChunkRange(begin, end, &first, &last)) {
if (last < first) return NS_ERROR_FAILURE;
if (last - first > MAX_CHUNK_RANGE) return NS_ERROR_FAILURE;
for (uint32_t num = first; num <= last; num++) {
if (aLine[0] == 'a') {
nsresult rv = mTableUpdate->NewAddExpiration(num);
if (NS_FAILED(rv)) {
return rv;
}
} else {
nsresult rv = mTableUpdate->NewSubExpiration(num);
if (NS_FAILED(rv)) {
return rv;
}
}
}
} else {
return NS_ERROR_FAILURE;
}
}
return NS_OK;
}
nsresult
ProtocolParserV2::ProcessChunkControl(const nsCString& aLine)
{
if (!mTableUpdate) {
NS_WARNING("Got a chunk before getting a table.");
return NS_ERROR_FAILURE;
}
mState = PROTOCOL_STATE_CHUNK;
char command;
mChunkState.Clear();
if (PR_sscanf(aLine.get(),
"%c:%d:%d:%d",
&command,
&mChunkState.num, &mChunkState.hashSize, &mChunkState.length)
!= 4)
{
NS_WARNING(("PR_sscanf failed"));
return NS_ERROR_FAILURE;
}
if (mChunkState.length > MAX_CHUNK_SIZE) {
NS_WARNING("Invalid length specified in update.");
return NS_ERROR_FAILURE;
}
if (!(mChunkState.hashSize == PREFIX_SIZE || mChunkState.hashSize == COMPLETE_SIZE)) {
NS_WARNING("Invalid hash size specified in update.");
return NS_ERROR_FAILURE;
}
if (StringEndsWith(mTableUpdate->TableName(),
NS_LITERAL_CSTRING("-shavar")) ||
StringEndsWith(mTableUpdate->TableName(),
NS_LITERAL_CSTRING("-simple"))) {
// Accommodate test tables ending in -simple for now.
mChunkState.type = (command == 'a') ? CHUNK_ADD : CHUNK_SUB;
} else if (StringEndsWith(mTableUpdate->TableName(),
NS_LITERAL_CSTRING("-digest256"))) {
mChunkState.type = (command == 'a') ? CHUNK_ADD_DIGEST : CHUNK_SUB_DIGEST;
}
nsresult rv;
switch (mChunkState.type) {
case CHUNK_ADD:
rv = mTableUpdate->NewAddChunk(mChunkState.num);
if (NS_FAILED(rv)) {
return rv;
}
break;
case CHUNK_SUB:
rv = mTableUpdate->NewSubChunk(mChunkState.num);
if (NS_FAILED(rv)) {
return rv;
}
break;
case CHUNK_ADD_DIGEST:
rv = mTableUpdate->NewAddChunk(mChunkState.num);
if (NS_FAILED(rv)) {
return rv;
}
break;
case CHUNK_SUB_DIGEST:
rv = mTableUpdate->NewSubChunk(mChunkState.num);
if (NS_FAILED(rv)) {
return rv;
}
break;
}
return NS_OK;
}
nsresult
ProtocolParserV2::ProcessForward(const nsCString& aLine)
{
const nsCSubstring &forward = Substring(aLine, 2);
return AddForward(forward);
}
nsresult
ProtocolParserV2::AddForward(const nsACString& aUrl)
{
if (!mTableUpdate) {
NS_WARNING("Forward without a table name.");
return NS_ERROR_FAILURE;
}
ForwardedUpdate *forward = mForwards.AppendElement();
forward->table = mTableUpdate->TableName();
forward->url.Assign(aUrl);
return NS_OK;
}
nsresult
ProtocolParserV2::ProcessChunk(bool* aDone)
{
if (!mTableUpdate) {
NS_WARNING("Processing chunk without an active table.");
return NS_ERROR_FAILURE;
}
NS_ASSERTION(mChunkState.num != 0, "Must have a chunk number.");
if (mPending.Length() < mChunkState.length) {
*aDone = true;
return NS_OK;
}
// Pull the chunk out of the pending stream data.
nsAutoCString chunk;
chunk.Assign(Substring(mPending, 0, mChunkState.length));
mPending.Cut(0, mChunkState.length);
*aDone = false;
mState = PROTOCOL_STATE_CONTROL;
if (StringEndsWith(mTableUpdate->TableName(),
NS_LITERAL_CSTRING("-shavar"))) {
return ProcessShaChunk(chunk);
}
if (StringEndsWith(mTableUpdate->TableName(),
NS_LITERAL_CSTRING("-digest256"))) {
return ProcessDigestChunk(chunk);
}
return ProcessPlaintextChunk(chunk);
}
/**
* Process a plaintext chunk (currently only used in unit tests).
*/
nsresult
ProtocolParserV2::ProcessPlaintextChunk(const nsACString& aChunk)
{
if (!mTableUpdate) {
NS_WARNING("Chunk received with no table.");
return NS_ERROR_FAILURE;
}
PARSER_LOG(("Handling a %d-byte simple chunk", aChunk.Length()));
nsTArray<nsCString> lines;
ParseString(PromiseFlatCString(aChunk), '\n', lines);
// non-hashed tables need to be hashed
for (uint32_t i = 0; i < lines.Length(); i++) {
nsCString& line = lines[i];
if (mChunkState.type == CHUNK_ADD) {
if (mChunkState.hashSize == COMPLETE_SIZE) {
Completion hash;
hash.FromPlaintext(line, mCryptoHash);
nsresult rv = mTableUpdate->NewAddComplete(mChunkState.num, hash);
if (NS_FAILED(rv)) {
return rv;
}
} else {
NS_ASSERTION(mChunkState.hashSize == 4, "Only 32- or 4-byte hashes can be used for add chunks.");
Prefix hash;
hash.FromPlaintext(line, mCryptoHash);
nsresult rv = mTableUpdate->NewAddPrefix(mChunkState.num, hash);
if (NS_FAILED(rv)) {
return rv;
}
}
} else {
nsCString::const_iterator begin, iter, end;
line.BeginReading(begin);
line.EndReading(end);
iter = begin;
uint32_t addChunk;
if (!FindCharInReadable(':', iter, end) ||
PR_sscanf(lines[i].get(), "%d:", &addChunk) != 1) {
NS_WARNING("Received sub chunk without associated add chunk.");
return NS_ERROR_FAILURE;
}
iter++;
if (mChunkState.hashSize == COMPLETE_SIZE) {
Completion hash;
hash.FromPlaintext(Substring(iter, end), mCryptoHash);
nsresult rv = mTableUpdate->NewSubComplete(addChunk, hash, mChunkState.num);
if (NS_FAILED(rv)) {
return rv;
}
} else {
NS_ASSERTION(mChunkState.hashSize == 4, "Only 32- or 4-byte hashes can be used for add chunks.");
Prefix hash;
hash.FromPlaintext(Substring(iter, end), mCryptoHash);
nsresult rv = mTableUpdate->NewSubPrefix(addChunk, hash, mChunkState.num);
if (NS_FAILED(rv)) {
return rv;
}
}
}
}
return NS_OK;
}
nsresult
ProtocolParserV2::ProcessShaChunk(const nsACString& aChunk)
{
uint32_t start = 0;
while (start < aChunk.Length()) {
// First four bytes are the domain key.
Prefix domain;
domain.Assign(Substring(aChunk, start, DOMAIN_SIZE));
start += DOMAIN_SIZE;
// Then a count of entries.
uint8_t numEntries = static_cast<uint8_t>(aChunk[start]);
start++;
PARSER_LOG(("Handling a %d-byte shavar chunk containing %u entries"
" for domain %X", aChunk.Length(), numEntries,
domain.ToUint32()));
nsresult rv;
if (mChunkState.type == CHUNK_ADD && mChunkState.hashSize == PREFIX_SIZE) {
rv = ProcessHostAdd(domain, numEntries, aChunk, &start);
} else if (mChunkState.type == CHUNK_ADD && mChunkState.hashSize == COMPLETE_SIZE) {
rv = ProcessHostAddComplete(numEntries, aChunk, &start);
} else if (mChunkState.type == CHUNK_SUB && mChunkState.hashSize == PREFIX_SIZE) {
rv = ProcessHostSub(domain, numEntries, aChunk, &start);
} else if (mChunkState.type == CHUNK_SUB && mChunkState.hashSize == COMPLETE_SIZE) {
rv = ProcessHostSubComplete(numEntries, aChunk, &start);
} else {
NS_WARNING("Unexpected chunk type/hash size!");
PARSER_LOG(("Got an unexpected chunk type/hash size: %s:%d",
mChunkState.type == CHUNK_ADD ? "add" : "sub",
mChunkState.hashSize));
return NS_ERROR_FAILURE;
}
NS_ENSURE_SUCCESS(rv, rv);
}
return NS_OK;
}
nsresult
ProtocolParserV2::ProcessDigestChunk(const nsACString& aChunk)
{
PARSER_LOG(("Handling a %d-byte digest256 chunk", aChunk.Length()));
if (mChunkState.type == CHUNK_ADD_DIGEST) {
return ProcessDigestAdd(aChunk);
}
if (mChunkState.type == CHUNK_SUB_DIGEST) {
return ProcessDigestSub(aChunk);
}
return NS_ERROR_UNEXPECTED;
}
nsresult
ProtocolParserV2::ProcessDigestAdd(const nsACString& aChunk)
{
// The ABNF format for add chunks is (HASH)+, where HASH is 32 bytes.
MOZ_ASSERT(aChunk.Length() % 32 == 0,
"Chunk length in bytes must be divisible by 4");
uint32_t start = 0;
while (start < aChunk.Length()) {
Completion hash;
hash.Assign(Substring(aChunk, start, COMPLETE_SIZE));
start += COMPLETE_SIZE;
nsresult rv = mTableUpdate->NewAddComplete(mChunkState.num, hash);
if (NS_FAILED(rv)) {
return rv;
}
}
return NS_OK;
}
nsresult
ProtocolParserV2::ProcessDigestSub(const nsACString& aChunk)
{
// The ABNF format for sub chunks is (ADDCHUNKNUM HASH)+, where ADDCHUNKNUM
// is a 4 byte chunk number, and HASH is 32 bytes.
MOZ_ASSERT(aChunk.Length() % 36 == 0,
"Chunk length in bytes must be divisible by 36");
uint32_t start = 0;
while (start < aChunk.Length()) {
// Read ADDCHUNKNUM
const nsCSubstring& addChunkStr = Substring(aChunk, start, 4);
start += 4;
uint32_t addChunk;
memcpy(&addChunk, addChunkStr.BeginReading(), 4);
addChunk = PR_ntohl(addChunk);
// Read the hash
Completion hash;
hash.Assign(Substring(aChunk, start, COMPLETE_SIZE));
start += COMPLETE_SIZE;
nsresult rv = mTableUpdate->NewSubComplete(addChunk, hash, mChunkState.num);
if (NS_FAILED(rv)) {
return rv;
}
}
return NS_OK;
}
nsresult
ProtocolParserV2::ProcessHostAdd(const Prefix& aDomain, uint8_t aNumEntries,
const nsACString& aChunk, uint32_t* aStart)
{
NS_ASSERTION(mChunkState.hashSize == PREFIX_SIZE,
"ProcessHostAdd should only be called for prefix hashes.");
if (aNumEntries == 0) {
nsresult rv = mTableUpdate->NewAddPrefix(mChunkState.num, aDomain);
if (NS_FAILED(rv)) {
return rv;
}
return NS_OK;
}
if (*aStart + (PREFIX_SIZE * aNumEntries) > aChunk.Length()) {
NS_WARNING("Chunk is not long enough to contain the expected entries.");
return NS_ERROR_FAILURE;
}
for (uint8_t i = 0; i < aNumEntries; i++) {
Prefix hash;
hash.Assign(Substring(aChunk, *aStart, PREFIX_SIZE));
PARSER_LOG(("Add prefix %X", hash.ToUint32()));
nsresult rv = mTableUpdate->NewAddPrefix(mChunkState.num, hash);
if (NS_FAILED(rv)) {
return rv;
}
*aStart += PREFIX_SIZE;
}
return NS_OK;
}
nsresult
ProtocolParserV2::ProcessHostSub(const Prefix& aDomain, uint8_t aNumEntries,
const nsACString& aChunk, uint32_t *aStart)
{
NS_ASSERTION(mChunkState.hashSize == PREFIX_SIZE,
"ProcessHostSub should only be called for prefix hashes.");
if (aNumEntries == 0) {
if ((*aStart) + 4 > aChunk.Length()) {
NS_WARNING("Received a zero-entry sub chunk without an associated add.");
return NS_ERROR_FAILURE;
}
const nsCSubstring& addChunkStr = Substring(aChunk, *aStart, 4);
*aStart += 4;
uint32_t addChunk;
memcpy(&addChunk, addChunkStr.BeginReading(), 4);
addChunk = PR_ntohl(addChunk);
PARSER_LOG(("Sub prefix (addchunk=%u)", addChunk));
nsresult rv = mTableUpdate->NewSubPrefix(addChunk, aDomain, mChunkState.num);
if (NS_FAILED(rv)) {
return rv;
}
return NS_OK;
}
if (*aStart + ((PREFIX_SIZE + 4) * aNumEntries) > aChunk.Length()) {
NS_WARNING("Chunk is not long enough to contain the expected entries.");
return NS_ERROR_FAILURE;
}
for (uint8_t i = 0; i < aNumEntries; i++) {
const nsCSubstring& addChunkStr = Substring(aChunk, *aStart, 4);
*aStart += 4;
uint32_t addChunk;
memcpy(&addChunk, addChunkStr.BeginReading(), 4);
addChunk = PR_ntohl(addChunk);
Prefix prefix;
prefix.Assign(Substring(aChunk, *aStart, PREFIX_SIZE));
*aStart += PREFIX_SIZE;
PARSER_LOG(("Sub prefix %X (addchunk=%u)", prefix.ToUint32(), addChunk));
nsresult rv = mTableUpdate->NewSubPrefix(addChunk, prefix, mChunkState.num);
if (NS_FAILED(rv)) {
return rv;
}
}
return NS_OK;
}
nsresult
ProtocolParserV2::ProcessHostAddComplete(uint8_t aNumEntries,
const nsACString& aChunk, uint32_t* aStart)
{
NS_ASSERTION(mChunkState.hashSize == COMPLETE_SIZE,
"ProcessHostAddComplete should only be called for complete hashes.");
if (aNumEntries == 0) {
// this is totally comprehensible.
// My sarcasm detector is going off!
NS_WARNING("Expected > 0 entries for a 32-byte hash add.");
return NS_OK;
}
if (*aStart + (COMPLETE_SIZE * aNumEntries) > aChunk.Length()) {
NS_WARNING("Chunk is not long enough to contain the expected entries.");
return NS_ERROR_FAILURE;
}
for (uint8_t i = 0; i < aNumEntries; i++) {
Completion hash;
hash.Assign(Substring(aChunk, *aStart, COMPLETE_SIZE));
nsresult rv = mTableUpdate->NewAddComplete(mChunkState.num, hash);
if (NS_FAILED(rv)) {
return rv;
}
*aStart += COMPLETE_SIZE;
}
return NS_OK;
}
nsresult
ProtocolParserV2::ProcessHostSubComplete(uint8_t aNumEntries,
const nsACString& aChunk, uint32_t* aStart)
{
NS_ASSERTION(mChunkState.hashSize == COMPLETE_SIZE,
"ProcessHostSubComplete should only be called for complete hashes.");
if (aNumEntries == 0) {
// this is totally comprehensible.
NS_WARNING("Expected > 0 entries for a 32-byte hash sub.");
return NS_OK;
}
if (*aStart + ((COMPLETE_SIZE + 4) * aNumEntries) > aChunk.Length()) {
NS_WARNING("Chunk is not long enough to contain the expected entries.");
return NS_ERROR_FAILURE;
}
for (uint8_t i = 0; i < aNumEntries; i++) {
Completion hash;
hash.Assign(Substring(aChunk, *aStart, COMPLETE_SIZE));
*aStart += COMPLETE_SIZE;
const nsCSubstring& addChunkStr = Substring(aChunk, *aStart, 4);
*aStart += 4;
uint32_t addChunk;
memcpy(&addChunk, addChunkStr.BeginReading(), 4);
addChunk = PR_ntohl(addChunk);
nsresult rv = mTableUpdate->NewSubComplete(addChunk, hash, mChunkState.num);
if (NS_FAILED(rv)) {
return rv;
}
}
return NS_OK;
}
bool
ProtocolParserV2::NextLine(nsACString& aLine)
{
int32_t newline = mPending.FindChar('\n');
if (newline == kNotFound) {
return false;
}
aLine.Assign(Substring(mPending, 0, newline));
mPending.Cut(0, newline + 1);
return true;
}
TableUpdate*
ProtocolParserV2::CreateTableUpdate(const nsACString& aTableName) const
{
return new TableUpdateV2(aTableName);
}
///////////////////////////////////////////////////////////////////////
// ProtocolParserProtobuf
ProtocolParserProtobuf::ProtocolParserProtobuf()
{
}
ProtocolParserProtobuf::~ProtocolParserProtobuf()
{
}
void
ProtocolParserProtobuf::SetCurrentTable(const nsACString& aTable)
{
// Should never occur.
MOZ_ASSERT_UNREACHABLE("SetCurrentTable shouldn't be called");
}
TableUpdate*
ProtocolParserProtobuf::CreateTableUpdate(const nsACString& aTableName) const
{
return new TableUpdateV4(aTableName);
}
nsresult
ProtocolParserProtobuf::AppendStream(const nsACString& aData)
{
// Protobuf data cannot be parsed progressively. Just save the incoming data.
mPending.Append(aData);
return NS_OK;
}
void
ProtocolParserProtobuf::End()
{
// mUpdateStatus will be updated to success as long as not all
// the responses are invalid.
mUpdateStatus = NS_ERROR_FAILURE;
FetchThreatListUpdatesResponse response;
if (!response.ParseFromArray(mPending.get(), mPending.Length())) {
NS_WARNING("ProtocolParserProtobuf failed parsing data.");
return;
}
for (int i = 0; i < response.list_update_responses_size(); i++) {
auto r = response.list_update_responses(i);
nsresult rv = ProcessOneResponse(r);
if (NS_SUCCEEDED(rv)) {
mUpdateStatus = rv;
} else {
NS_WARNING("Failed to process one response.");
}
}
}
nsresult
ProtocolParserProtobuf::ProcessOneResponse(const ListUpdateResponse& aResponse)
{
// A response must have a threat type.
if (!aResponse.has_threat_type()) {
NS_WARNING("Threat type not initialized. This seems to be an invalid response.");
return NS_ERROR_FAILURE;
}
// Convert threat type to list name.
nsCOMPtr<nsIUrlClassifierUtils> urlUtil =
do_GetService(NS_URLCLASSIFIERUTILS_CONTRACTID);
nsCString possibleListNames;
nsresult rv = urlUtil->ConvertThreatTypeToListNames(aResponse.threat_type(),
possibleListNames);
if (NS_FAILED(rv)) {
PARSER_LOG((nsPrintfCString("Threat type to list name conversion error: %d",
aResponse.threat_type())).get());
return NS_ERROR_FAILURE;
}
// Match the table name we received with one of the ones we requested.
// We ignore the case where a threat type matches more than one list
// per provider and return the first one. See bug 1287059."
nsCString listName;
nsTArray<nsCString> possibleListNameArray;
Classifier::SplitTables(possibleListNames, possibleListNameArray);
for (auto possibleName : possibleListNameArray) {
if (mRequestedTables.Contains(possibleName)) {
listName = possibleName;
break;
}
}
if (listName.IsEmpty()) {
PARSER_LOG(("We received an update for a list we didn't ask for. Ignoring it."));
return NS_ERROR_FAILURE;
}
// Test if this is a full update.
bool isFullUpdate = false;
if (aResponse.has_response_type()) {
isFullUpdate =
aResponse.response_type() == ListUpdateResponse::FULL_UPDATE;
} else {
NS_WARNING("Response type not initialized.");
return NS_ERROR_FAILURE;
}
// Warn if there's no new state.
if (!aResponse.has_new_client_state()) {
NS_WARNING("New state not initialized.");
return NS_ERROR_FAILURE;
}
auto tu = GetTableUpdate(nsCString(listName.get()));
auto tuV4 = TableUpdate::Cast<TableUpdateV4>(tu);
NS_ENSURE_TRUE(tuV4, NS_ERROR_FAILURE);
nsCString state(aResponse.new_client_state().c_str(),
aResponse.new_client_state().size());
tuV4->SetNewClientState(state);
if (aResponse.has_checksum()) {
tuV4->NewChecksum(aResponse.checksum().sha256());
}
PARSER_LOG(("==== Update for threat type '%d' ====", aResponse.threat_type()));
PARSER_LOG(("* listName: %s\n", listName.get()));
PARSER_LOG(("* newState: %s\n", aResponse.new_client_state().c_str()));
PARSER_LOG(("* isFullUpdate: %s\n", (isFullUpdate ? "yes" : "no")));
PARSER_LOG(("* hasChecksum: %s\n", (aResponse.has_checksum() ? "yes" : "no")));
tuV4->SetFullUpdate(isFullUpdate);
ProcessAdditionOrRemoval(*tuV4, aResponse.additions(), true /*aIsAddition*/);
ProcessAdditionOrRemoval(*tuV4, aResponse.removals(), false);
PARSER_LOG(("\n\n"));
return NS_OK;
}
nsresult
ProtocolParserProtobuf::ProcessAdditionOrRemoval(TableUpdateV4& aTableUpdate,
const ThreatEntrySetList& aUpdate,
bool aIsAddition)
{
nsresult ret = NS_OK;
for (int i = 0; i < aUpdate.size(); i++) {
auto update = aUpdate.Get(i);
if (!update.has_compression_type()) {
NS_WARNING(nsPrintfCString("%s with no compression type.",
aIsAddition ? "Addition" : "Removal").get());
continue;
}
switch (update.compression_type()) {
case COMPRESSION_TYPE_UNSPECIFIED:
NS_WARNING("Unspecified compression type.");
break;
case RAW:
ret = (aIsAddition ? ProcessRawAddition(aTableUpdate, update)
: ProcessRawRemoval(aTableUpdate, update));
break;
case RICE:
ret = (aIsAddition ? ProcessEncodedAddition(aTableUpdate, update)
: ProcessEncodedRemoval(aTableUpdate, update));
break;
}
}
return ret;
}
nsresult
ProtocolParserProtobuf::ProcessRawAddition(TableUpdateV4& aTableUpdate,
const ThreatEntrySet& aAddition)
{
if (!aAddition.has_raw_hashes()) {
PARSER_LOG(("* No raw addition."));
return NS_OK;
}
auto rawHashes = aAddition.raw_hashes();
if (!rawHashes.has_prefix_size()) {
NS_WARNING("Raw hash has no prefix size");
return NS_OK;
}
auto prefixes = rawHashes.raw_hashes();
if (4 == rawHashes.prefix_size()) {
// Process fixed length prefixes separately.
uint32_t* fixedLengthPrefixes = (uint32_t*)prefixes.c_str();
size_t numOfFixedLengthPrefixes = prefixes.size() / 4;
PARSER_LOG(("* Raw addition (4 bytes)"));
PARSER_LOG((" - # of prefixes: %d", numOfFixedLengthPrefixes));
PARSER_LOG((" - Memory address: 0x%p", fixedLengthPrefixes));
} else {
// TODO: Process variable length prefixes including full hashes.
// See Bug 1283009.
PARSER_LOG((" Raw addition (%d bytes)", rawHashes.prefix_size()));
}
if (!rawHashes.mutable_raw_hashes()) {
PARSER_LOG(("Unable to get mutable raw hashes. Can't perform a string move."));
return NS_ERROR_FAILURE;
}
aTableUpdate.NewPrefixes(rawHashes.prefix_size(),
*rawHashes.mutable_raw_hashes());
return NS_OK;
}
nsresult
ProtocolParserProtobuf::ProcessRawRemoval(TableUpdateV4& aTableUpdate,
const ThreatEntrySet& aRemoval)
{
if (!aRemoval.has_raw_indices()) {
NS_WARNING("A removal has no indices.");
return NS_OK;
}
// indices is an array of int32.
auto indices = aRemoval.raw_indices().indices();
PARSER_LOG(("* Raw removal"));
PARSER_LOG((" - # of removal: %d", indices.size()));
aTableUpdate.NewRemovalIndices((const uint32_t*)indices.data(),
indices.size());
return NS_OK;
}
static nsresult
DoRiceDeltaDecode(const RiceDeltaEncoding& aEncoding,
nsTArray<uint32_t>& aDecoded)
{
// Sanity check of the encoding info.
if (!aEncoding.has_first_value() ||
!aEncoding.has_rice_parameter() ||
!aEncoding.has_num_entries() ||
!aEncoding.has_encoded_data()) {
PARSER_LOG(("The encoding info is incomplete."));
return NS_ERROR_FAILURE;
}
PARSER_LOG(("* Encoding info:"));
PARSER_LOG((" - First value: %d", aEncoding.first_value()));
PARSER_LOG((" - Num of entries: %d", aEncoding.num_entries()));
PARSER_LOG((" - Rice parameter: %d", aEncoding.rice_parameter()));
// Set up the input buffer. Note that the bits should be read
// from LSB to MSB so that we in-place reverse the bits before
// feeding to the decoder.
auto encoded = const_cast<RiceDeltaEncoding&>(aEncoding).mutable_encoded_data();
RiceDeltaDecoder decoder((uint8_t*)encoded->c_str(), encoded->size());
// Setup the output buffer. The "first value" is included in
// the output buffer.
aDecoded.SetLength(aEncoding.num_entries() + 1);
aDecoded[0] = aEncoding.first_value();
// Decode!
bool rv = decoder.Decode(aEncoding.rice_parameter(),
aEncoding.first_value(), // first value.
aEncoding.num_entries(), // # of entries (first value not included).
&aDecoded[1]);
NS_ENSURE_TRUE(rv, NS_ERROR_FAILURE);
return NS_OK;
}
nsresult
ProtocolParserProtobuf::ProcessEncodedAddition(TableUpdateV4& aTableUpdate,
const ThreatEntrySet& aAddition)
{
if (!aAddition.has_rice_hashes()) {
PARSER_LOG(("* No rice encoded addition."));
return NS_OK;
}
nsTArray<uint32_t> decoded;
nsresult rv = DoRiceDeltaDecode(aAddition.rice_hashes(), decoded);
NS_ENSURE_SUCCESS(rv, rv);
// Say we have the following raw prefixes
// BE LE
// 00 00 00 01 1 16777216
// 00 00 02 00 512 131072
// 00 03 00 00 196608 768
// 04 00 00 00 67108864 4
//
// which can be treated as uint32 (big-endian) sorted in increasing order:
//
// [1, 512, 196608, 67108864]
//
// According to https://developers.google.com/safe-browsing/v4/compression,
// the following should be done prior to compression:
//
// 1) re-interpret in little-endian ==> [16777216, 131072, 768, 4]
// 2) sort in increasing order ==> [4, 768, 131072, 16777216]
//
// In order to get the original byte stream from |decoded|
// ([4, 768, 131072, 16777216] in this case), we have to:
//
// 1) sort in big-endian order ==> [16777216, 131072, 768, 4]
// 2) copy each uint32 in little-endian to the result string
//
// The 4-byte prefixes have to be re-sorted in Big-endian increasing order.
struct CompareBigEndian
{
bool Equals(const uint32_t& aA, const uint32_t& aB) const
{
return aA == aB;
}
bool LessThan(const uint32_t& aA, const uint32_t& aB) const
{
return NativeEndian::swapToBigEndian(aA) <
NativeEndian::swapToBigEndian(aB);
}
};
decoded.Sort(CompareBigEndian());
// The encoded prefixes are always 4 bytes.
std::string prefixes;
for (size_t i = 0; i < decoded.Length(); i++) {
// Note that the third argument is the number of elements we want
// to copy (and swap) but not the number of bytes we want to copy.
char p[4];
NativeEndian::copyAndSwapToLittleEndian(p, &decoded[i], 1);
prefixes.append(p, 4);
}
aTableUpdate.NewPrefixes(4, prefixes);
return NS_OK;
}
nsresult
ProtocolParserProtobuf::ProcessEncodedRemoval(TableUpdateV4& aTableUpdate,
const ThreatEntrySet& aRemoval)
{
if (!aRemoval.has_rice_indices()) {
PARSER_LOG(("* No rice encoded removal."));
return NS_OK;
}
nsTArray<uint32_t> decoded;
nsresult rv = DoRiceDeltaDecode(aRemoval.rice_indices(), decoded);
NS_ENSURE_SUCCESS(rv, rv);
// The encoded prefixes are always 4 bytes.
aTableUpdate.NewRemovalIndices(&decoded[0], decoded.Length());
return NS_OK;
}
} // namespace safebrowsing
} // namespace mozilla