gecko-dev/netwerk/base/BackgroundFileSaver.cpp

1126 строки
39 KiB
C++

/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
/* vim: set ts=2 et sw=2 tw=80: */
/* 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 "BackgroundFileSaver.h"
#include "ScopedNSSTypes.h"
#include "mozilla/ArrayAlgorithm.h"
#include "mozilla/Casting.h"
#include "mozilla/Logging.h"
#include "mozilla/Telemetry.h"
#include "nsCOMArray.h"
#include "nsComponentManagerUtils.h"
#include "nsDependentSubstring.h"
#include "nsIAsyncInputStream.h"
#include "nsIFile.h"
#include "nsIMutableArray.h"
#include "nsIPipe.h"
#include "nsNetUtil.h"
#include "nsThreadUtils.h"
#include "pk11pub.h"
#include "secoidt.h"
#ifdef XP_WIN
# include <windows.h>
# include <softpub.h>
# include <wintrust.h>
#endif // XP_WIN
namespace mozilla {
namespace net {
// MOZ_LOG=BackgroundFileSaver:5
static LazyLogModule prlog("BackgroundFileSaver");
#define LOG(args) MOZ_LOG(prlog, mozilla::LogLevel::Debug, args)
#define LOG_ENABLED() MOZ_LOG_TEST(prlog, mozilla::LogLevel::Debug)
////////////////////////////////////////////////////////////////////////////////
//// Globals
/**
* Buffer size for writing to the output file or reading from the input file.
*/
#define BUFFERED_IO_SIZE (1024 * 32)
/**
* When this upper limit is reached, the original request is suspended.
*/
#define REQUEST_SUSPEND_AT (1024 * 1024 * 4)
/**
* When this lower limit is reached, the original request is resumed.
*/
#define REQUEST_RESUME_AT (1024 * 1024 * 2)
////////////////////////////////////////////////////////////////////////////////
//// NotifyTargetChangeRunnable
/**
* Runnable object used to notify the control thread that file contents will now
* be saved to the specified file.
*/
class NotifyTargetChangeRunnable final : public Runnable {
public:
NotifyTargetChangeRunnable(BackgroundFileSaver* aSaver, nsIFile* aTarget)
: Runnable("net::NotifyTargetChangeRunnable"),
mSaver(aSaver),
mTarget(aTarget) {}
NS_IMETHOD Run() override { return mSaver->NotifyTargetChange(mTarget); }
private:
RefPtr<BackgroundFileSaver> mSaver;
nsCOMPtr<nsIFile> mTarget;
};
////////////////////////////////////////////////////////////////////////////////
//// BackgroundFileSaver
uint32_t BackgroundFileSaver::sThreadCount = 0;
uint32_t BackgroundFileSaver::sTelemetryMaxThreadCount = 0;
BackgroundFileSaver::BackgroundFileSaver()
: mControlEventTarget(nullptr),
mBackgroundET(nullptr),
mPipeOutputStream(nullptr),
mPipeInputStream(nullptr),
mObserver(nullptr),
mLock("BackgroundFileSaver.mLock"),
mWorkerThreadAttentionRequested(false),
mFinishRequested(false),
mComplete(false),
mStatus(NS_OK),
mAppend(false),
mInitialTarget(nullptr),
mInitialTargetKeepPartial(false),
mRenamedTarget(nullptr),
mRenamedTargetKeepPartial(false),
mAsyncCopyContext(nullptr),
mSha256Enabled(false),
mSignatureInfoEnabled(false),
mActualTarget(nullptr),
mActualTargetKeepPartial(false),
mDigestContext(nullptr) {
LOG(("Created BackgroundFileSaver [this = %p]", this));
}
BackgroundFileSaver::~BackgroundFileSaver() {
LOG(("Destroying BackgroundFileSaver [this = %p]", this));
}
// Called on the control thread.
nsresult BackgroundFileSaver::Init() {
MOZ_ASSERT(NS_IsMainThread(), "This should be called on the main thread");
nsresult rv;
rv = NS_NewPipe2(getter_AddRefs(mPipeInputStream),
getter_AddRefs(mPipeOutputStream), true, true, 0,
HasInfiniteBuffer() ? UINT32_MAX : 0);
NS_ENSURE_SUCCESS(rv, rv);
mControlEventTarget = GetCurrentEventTarget();
NS_ENSURE_TRUE(mControlEventTarget, NS_ERROR_NOT_INITIALIZED);
rv = NS_CreateBackgroundTaskQueue("BgFileSaver",
getter_AddRefs(mBackgroundET));
NS_ENSURE_SUCCESS(rv, rv);
sThreadCount++;
if (sThreadCount > sTelemetryMaxThreadCount) {
sTelemetryMaxThreadCount = sThreadCount;
}
return NS_OK;
}
// Called on the control thread.
NS_IMETHODIMP
BackgroundFileSaver::GetObserver(nsIBackgroundFileSaverObserver** aObserver) {
NS_ENSURE_ARG_POINTER(aObserver);
*aObserver = mObserver;
NS_IF_ADDREF(*aObserver);
return NS_OK;
}
// Called on the control thread.
NS_IMETHODIMP
BackgroundFileSaver::SetObserver(nsIBackgroundFileSaverObserver* aObserver) {
mObserver = aObserver;
return NS_OK;
}
// Called on the control thread.
NS_IMETHODIMP
BackgroundFileSaver::EnableAppend() {
MOZ_ASSERT(NS_IsMainThread(), "This should be called on the main thread");
MutexAutoLock lock(mLock);
mAppend = true;
return NS_OK;
}
// Called on the control thread.
NS_IMETHODIMP
BackgroundFileSaver::SetTarget(nsIFile* aTarget, bool aKeepPartial) {
NS_ENSURE_ARG(aTarget);
{
MutexAutoLock lock(mLock);
if (!mInitialTarget) {
aTarget->Clone(getter_AddRefs(mInitialTarget));
mInitialTargetKeepPartial = aKeepPartial;
} else {
aTarget->Clone(getter_AddRefs(mRenamedTarget));
mRenamedTargetKeepPartial = aKeepPartial;
}
}
// After the worker thread wakes up because attention is requested, it will
// rename or create the target file as requested, and start copying data.
return GetWorkerThreadAttention(true);
}
// Called on the control thread.
NS_IMETHODIMP
BackgroundFileSaver::Finish(nsresult aStatus) {
nsresult rv;
// This will cause the NS_AsyncCopy operation, if it's in progress, to consume
// all the data that is still in the pipe, and then finish.
rv = mPipeOutputStream->Close();
NS_ENSURE_SUCCESS(rv, rv);
// Ensure that, when we get attention from the worker thread, if no pending
// rename operation is waiting, the operation will complete.
{
MutexAutoLock lock(mLock);
mFinishRequested = true;
if (NS_SUCCEEDED(mStatus)) {
mStatus = aStatus;
}
}
// After the worker thread wakes up because attention is requested, it will
// process the completion conditions, detect that completion is requested, and
// notify the main thread of the completion. If this function was called with
// a success code, we wait for the copy to finish before processing the
// completion conditions, otherwise we interrupt the copy immediately.
return GetWorkerThreadAttention(NS_FAILED(aStatus));
}
NS_IMETHODIMP
BackgroundFileSaver::EnableSha256() {
MOZ_ASSERT(NS_IsMainThread(),
"Can't enable sha256 or initialize NSS off the main thread");
// Ensure Personal Security Manager is initialized. This is required for
// PK11_* operations to work.
nsresult rv;
nsCOMPtr<nsISupports> nssDummy = do_GetService("@mozilla.org/psm;1", &rv);
NS_ENSURE_SUCCESS(rv, rv);
mSha256Enabled = true;
return NS_OK;
}
NS_IMETHODIMP
BackgroundFileSaver::GetSha256Hash(nsACString& aHash) {
MOZ_ASSERT(NS_IsMainThread(), "Can't inspect sha256 off the main thread");
// We acquire a lock because mSha256 is written on the worker thread.
MutexAutoLock lock(mLock);
if (mSha256.IsEmpty()) {
return NS_ERROR_NOT_AVAILABLE;
}
aHash = mSha256;
return NS_OK;
}
NS_IMETHODIMP
BackgroundFileSaver::EnableSignatureInfo() {
MOZ_ASSERT(NS_IsMainThread(),
"Can't enable signature extraction off the main thread");
// Ensure Personal Security Manager is initialized.
nsresult rv;
nsCOMPtr<nsISupports> nssDummy = do_GetService("@mozilla.org/psm;1", &rv);
NS_ENSURE_SUCCESS(rv, rv);
mSignatureInfoEnabled = true;
return NS_OK;
}
NS_IMETHODIMP
BackgroundFileSaver::GetSignatureInfo(
nsTArray<nsTArray<nsTArray<uint8_t>>>& aSignatureInfo) {
MOZ_ASSERT(NS_IsMainThread(), "Can't inspect signature off the main thread");
// We acquire a lock because mSignatureInfo is written on the worker thread.
MutexAutoLock lock(mLock);
if (!mComplete || !mSignatureInfoEnabled) {
return NS_ERROR_NOT_AVAILABLE;
}
for (const auto& signatureChain : mSignatureInfo) {
aSignatureInfo.AppendElement(TransformIntoNewArray(
signatureChain, [](const auto& element) { return element.Clone(); }));
}
return NS_OK;
}
// Called on the control thread.
nsresult BackgroundFileSaver::GetWorkerThreadAttention(
bool aShouldInterruptCopy) {
nsresult rv;
MutexAutoLock lock(mLock);
// We only require attention one time. If this function is called two times
// before the worker thread wakes up, and the first has aShouldInterruptCopy
// false and the second true, we won't forcibly interrupt the copy from the
// control thread. However, that never happens, because calling Finish with a
// success code is the only case that may result in aShouldInterruptCopy being
// false. In that case, we won't call this function again, because consumers
// should not invoke other methods on the control thread after calling Finish.
// And in any case, Finish already closes one end of the pipe, causing the
// copy to finish properly on its own.
if (mWorkerThreadAttentionRequested) {
return NS_OK;
}
if (!mAsyncCopyContext) {
// Background event queues are not shutdown and could be called after
// the queue is reset to null. To match the behavior of nsIThread
// return NS_ERROR_UNEXPECTED
if (!mBackgroundET) {
return NS_ERROR_UNEXPECTED;
}
// Copy is not in progress, post an event to handle the change manually.
rv = mBackgroundET->Dispatch(
NewRunnableMethod("net::BackgroundFileSaver::ProcessAttention", this,
&BackgroundFileSaver::ProcessAttention),
NS_DISPATCH_EVENT_MAY_BLOCK);
NS_ENSURE_SUCCESS(rv, rv);
} else if (aShouldInterruptCopy) {
// Interrupt the copy. The copy will be resumed, if needed, by the
// ProcessAttention function, invoked by the AsyncCopyCallback function.
NS_CancelAsyncCopy(mAsyncCopyContext, NS_ERROR_ABORT);
}
// Indicate that attention has been requested successfully, there is no need
// to post another event until the worker thread processes the current one.
mWorkerThreadAttentionRequested = true;
return NS_OK;
}
// Called on the worker thread.
// static
void BackgroundFileSaver::AsyncCopyCallback(void* aClosure, nsresult aStatus) {
// We called NS_ADDREF_THIS when NS_AsyncCopy started, to keep the object
// alive even if other references disappeared. At the end of this method,
// we've finished using the object and can safely release our reference.
RefPtr<BackgroundFileSaver> self =
dont_AddRef((BackgroundFileSaver*)aClosure);
{
MutexAutoLock lock(self->mLock);
// Now that the copy was interrupted or terminated, any notification from
// the control thread requires an event to be posted to the worker thread.
self->mAsyncCopyContext = nullptr;
// When detecting failures, ignore the status code we use to interrupt.
if (NS_FAILED(aStatus) && aStatus != NS_ERROR_ABORT &&
NS_SUCCEEDED(self->mStatus)) {
self->mStatus = aStatus;
}
}
(void)self->ProcessAttention();
}
// Called on the worker thread.
nsresult BackgroundFileSaver::ProcessAttention() {
nsresult rv;
// This function is called whenever the attention of the worker thread has
// been requested. This may happen in these cases:
// * We are about to start the copy for the first time. In this case, we are
// called from an event posted on the worker thread from the control thread
// by GetWorkerThreadAttention, and mAsyncCopyContext is null.
// * We have interrupted the copy for some reason. In this case, we are
// called by AsyncCopyCallback, and mAsyncCopyContext is null.
// * We are currently executing ProcessStateChange, and attention is requested
// by the control thread, for example because SetTarget or Finish have been
// called. In this case, we are called from from an event posted through
// GetWorkerThreadAttention. While mAsyncCopyContext was always null when
// the event was posted, at this point mAsyncCopyContext may not be null
// anymore, because ProcessStateChange may have started the copy before the
// event that called this function was processed on the worker thread.
// If mAsyncCopyContext is not null, we interrupt the copy and re-enter
// through AsyncCopyCallback. This allows us to check if, for instance, we
// should rename the target file. We will then restart the copy if needed.
if (mAsyncCopyContext) {
NS_CancelAsyncCopy(mAsyncCopyContext, NS_ERROR_ABORT);
return NS_OK;
}
// Use the current shared state to determine the next operation to execute.
rv = ProcessStateChange();
if (NS_FAILED(rv)) {
// If something failed while processing, terminate the operation now.
{
MutexAutoLock lock(mLock);
if (NS_SUCCEEDED(mStatus)) {
mStatus = rv;
}
}
// Ensure we notify completion now that the operation failed.
CheckCompletion();
}
return NS_OK;
}
// Called on the worker thread.
nsresult BackgroundFileSaver::ProcessStateChange() {
nsresult rv;
// We might have been notified because the operation is complete, verify.
if (CheckCompletion()) {
return NS_OK;
}
// Get a copy of the current shared state for the worker thread.
nsCOMPtr<nsIFile> initialTarget;
bool initialTargetKeepPartial;
nsCOMPtr<nsIFile> renamedTarget;
bool renamedTargetKeepPartial;
bool sha256Enabled;
bool append;
{
MutexAutoLock lock(mLock);
initialTarget = mInitialTarget;
initialTargetKeepPartial = mInitialTargetKeepPartial;
renamedTarget = mRenamedTarget;
renamedTargetKeepPartial = mRenamedTargetKeepPartial;
sha256Enabled = mSha256Enabled;
append = mAppend;
// From now on, another attention event needs to be posted if state changes.
mWorkerThreadAttentionRequested = false;
}
// The initial target can only be null if it has never been assigned. In this
// case, there is nothing to do since we never created any output file.
if (!initialTarget) {
return NS_OK;
}
// Determine if we are processing the attention request for the first time.
bool isContinuation = !!mActualTarget;
if (!isContinuation) {
// Assign the target file for the first time.
mActualTarget = initialTarget;
mActualTargetKeepPartial = initialTargetKeepPartial;
}
// Verify whether we have actually been instructed to use a different file.
// This may happen the first time this function is executed, if SetTarget was
// called two times before the worker thread processed the attention request.
bool equalToCurrent = false;
if (renamedTarget) {
rv = mActualTarget->Equals(renamedTarget, &equalToCurrent);
NS_ENSURE_SUCCESS(rv, rv);
if (!equalToCurrent) {
// If we were asked to rename the file but the initial file did not exist,
// we simply create the file in the renamed location. We avoid this check
// if we have already started writing the output file ourselves.
bool exists = true;
if (!isContinuation) {
rv = mActualTarget->Exists(&exists);
NS_ENSURE_SUCCESS(rv, rv);
}
if (exists) {
// We are moving the previous target file to a different location.
nsCOMPtr<nsIFile> renamedTargetParentDir;
rv = renamedTarget->GetParent(getter_AddRefs(renamedTargetParentDir));
NS_ENSURE_SUCCESS(rv, rv);
nsAutoString renamedTargetName;
rv = renamedTarget->GetLeafName(renamedTargetName);
NS_ENSURE_SUCCESS(rv, rv);
// We must delete any existing target file before moving the current
// one.
rv = renamedTarget->Exists(&exists);
NS_ENSURE_SUCCESS(rv, rv);
if (exists) {
rv = renamedTarget->Remove(false);
NS_ENSURE_SUCCESS(rv, rv);
}
// Move the file. If this fails, we still reference the original file
// in mActualTarget, so that it is deleted if requested. If this
// succeeds, the nsIFile instance referenced by mActualTarget mutates
// and starts pointing to the new file, but we'll discard the reference.
rv = mActualTarget->MoveTo(renamedTargetParentDir, renamedTargetName);
NS_ENSURE_SUCCESS(rv, rv);
}
// We should not only update the mActualTarget with renameTarget when
// they point to the different files.
// In this way, if mActualTarget and renamedTarget point to the same file
// with different addresses, "CheckCompletion()" will return false
// forever.
}
// Update mActualTarget with renameTarget,
// even if they point to the same file.
mActualTarget = renamedTarget;
mActualTargetKeepPartial = renamedTargetKeepPartial;
}
// Notify if the target file name actually changed.
if (!equalToCurrent) {
// We must clone the nsIFile instance because mActualTarget is not
// immutable, it may change if the target is renamed later.
nsCOMPtr<nsIFile> actualTargetToNotify;
rv = mActualTarget->Clone(getter_AddRefs(actualTargetToNotify));
NS_ENSURE_SUCCESS(rv, rv);
RefPtr<NotifyTargetChangeRunnable> event =
new NotifyTargetChangeRunnable(this, actualTargetToNotify);
NS_ENSURE_TRUE(event, NS_ERROR_FAILURE);
rv = mControlEventTarget->Dispatch(event, NS_DISPATCH_NORMAL);
NS_ENSURE_SUCCESS(rv, rv);
}
if (isContinuation) {
// The pending rename operation might be the last task before finishing. We
// may return here only if we have already created the target file.
if (CheckCompletion()) {
return NS_OK;
}
// Even if the operation did not complete, the pipe input stream may be
// empty and may have been closed already. We detect this case using the
// Available property, because it never returns an error if there is more
// data to be consumed. If the pipe input stream is closed, we just exit
// and wait for more calls like SetTarget or Finish to be invoked on the
// control thread. However, we still truncate the file or create the
// initial digest context if we are expected to do that.
uint64_t available;
rv = mPipeInputStream->Available(&available);
if (NS_FAILED(rv)) {
return NS_OK;
}
}
// Create the digest context if requested and NSS hasn't been shut down.
if (sha256Enabled && !mDigestContext) {
mDigestContext =
UniquePK11Context(PK11_CreateDigestContext(SEC_OID_SHA256));
NS_ENSURE_TRUE(mDigestContext, NS_ERROR_OUT_OF_MEMORY);
}
// When we are requested to append to an existing file, we should read the
// existing data and ensure we include it as part of the final hash.
if (mDigestContext && append && !isContinuation) {
nsCOMPtr<nsIInputStream> inputStream;
rv = NS_NewLocalFileInputStream(getter_AddRefs(inputStream), mActualTarget,
PR_RDONLY | nsIFile::OS_READAHEAD);
if (rv != NS_ERROR_FILE_NOT_FOUND) {
NS_ENSURE_SUCCESS(rv, rv);
char buffer[BUFFERED_IO_SIZE];
while (true) {
uint32_t count;
rv = inputStream->Read(buffer, BUFFERED_IO_SIZE, &count);
NS_ENSURE_SUCCESS(rv, rv);
if (count == 0) {
// We reached the end of the file.
break;
}
nsresult rv = MapSECStatus(
PK11_DigestOp(mDigestContext.get(),
BitwiseCast<unsigned char*, char*>(buffer), count));
NS_ENSURE_SUCCESS(rv, rv);
}
rv = inputStream->Close();
NS_ENSURE_SUCCESS(rv, rv);
}
}
// We will append to the initial target file only if it was requested by the
// caller, but we'll always append on subsequent accesses to the target file.
int32_t creationIoFlags;
if (isContinuation) {
creationIoFlags = PR_APPEND;
} else {
creationIoFlags = (append ? PR_APPEND : PR_TRUNCATE) | PR_CREATE_FILE;
}
// Create the target file, or append to it if we already started writing it.
// The 0600 permissions are used while the file is being downloaded, and for
// interrupted downloads. Those may be located in the system temporary
// directory, as well as the target directory, and generally have a ".part"
// extension. Those part files should never be group or world-writable even
// if the umask allows it.
nsCOMPtr<nsIOutputStream> outputStream;
rv = NS_NewLocalFileOutputStream(getter_AddRefs(outputStream), mActualTarget,
PR_WRONLY | creationIoFlags, 0600);
NS_ENSURE_SUCCESS(rv, rv);
nsCOMPtr<nsIOutputStream> bufferedStream;
rv = NS_NewBufferedOutputStream(getter_AddRefs(bufferedStream),
outputStream.forget(), BUFFERED_IO_SIZE);
NS_ENSURE_SUCCESS(rv, rv);
outputStream = bufferedStream;
// Wrap the output stream so that it feeds the digest context if needed.
if (mDigestContext) {
// Constructing the DigestOutputStream cannot fail. Passing mDigestContext
// to DigestOutputStream is safe, because BackgroundFileSaver always
// outlives the outputStream. BackgroundFileSaver is reference-counted
// before the call to AsyncCopy, and mDigestContext is never destroyed
// before AsyncCopyCallback.
outputStream = new DigestOutputStream(outputStream, mDigestContext.get());
}
// Start copying our input to the target file. No errors can be raised past
// this point if the copy starts, since they should be handled by the thread.
{
MutexAutoLock lock(mLock);
rv = NS_AsyncCopy(mPipeInputStream, outputStream, mBackgroundET,
NS_ASYNCCOPY_VIA_READSEGMENTS, 4096, AsyncCopyCallback,
this, false, true, getter_AddRefs(mAsyncCopyContext),
GetProgressCallback());
if (NS_FAILED(rv)) {
NS_WARNING("NS_AsyncCopy failed.");
mAsyncCopyContext = nullptr;
return rv;
}
}
// If the operation succeeded, we must ensure that we keep this object alive
// for the entire duration of the copy, since only the raw pointer will be
// provided as the argument of the AsyncCopyCallback function. We can add the
// reference now, after NS_AsyncCopy returned, because it always starts
// processing asynchronously, and there is no risk that the callback is
// invoked before we reach this point. If the operation failed instead, then
// AsyncCopyCallback will never be called.
NS_ADDREF_THIS();
return NS_OK;
}
// Called on the worker thread.
bool BackgroundFileSaver::CheckCompletion() {
nsresult rv;
MOZ_ASSERT(!mAsyncCopyContext,
"Should not be copying when checking completion conditions.");
bool failed = true;
{
MutexAutoLock lock(mLock);
if (mComplete) {
return true;
}
// If an error occurred, we don't need to do the checks in this code block,
// and the operation can be completed immediately with a failure code.
if (NS_SUCCEEDED(mStatus)) {
failed = false;
// We did not incur in an error, so we must determine if we can stop now.
// If the Finish method has not been called, we can just continue now.
if (!mFinishRequested) {
return false;
}
// We can only stop when all the operations requested by the control
// thread have been processed. First, we check whether we have processed
// the first SetTarget call, if any. Then, we check whether we have
// processed any rename requested by subsequent SetTarget calls.
if ((mInitialTarget && !mActualTarget) ||
(mRenamedTarget && mRenamedTarget != mActualTarget)) {
return false;
}
// If we still have data to write to the output file, allow the copy
// operation to resume. The Available getter may return an error if one
// of the pipe's streams has been already closed.
uint64_t available;
rv = mPipeInputStream->Available(&available);
if (NS_SUCCEEDED(rv) && available != 0) {
return false;
}
}
mComplete = true;
}
// Ensure we notify completion now that the operation finished.
// Do a best-effort attempt to remove the file if required.
if (failed && mActualTarget && !mActualTargetKeepPartial) {
(void)mActualTarget->Remove(false);
}
// Finish computing the hash
if (!failed && mDigestContext) {
Digest d;
rv = d.End(SEC_OID_SHA256, mDigestContext);
if (NS_SUCCEEDED(rv)) {
MutexAutoLock lock(mLock);
mSha256 = nsDependentCSubstring(
BitwiseCast<char*, unsigned char*>(d.get().data), d.get().len);
}
}
// Compute the signature of the binary. ExtractSignatureInfo doesn't do
// anything on non-Windows platforms except return an empty nsIArray.
if (!failed && mActualTarget) {
nsString filePath;
mActualTarget->GetTarget(filePath);
nsresult rv = ExtractSignatureInfo(filePath);
if (NS_FAILED(rv)) {
LOG(("Unable to extract signature information [this = %p].", this));
} else {
LOG(("Signature extraction success! [this = %p]", this));
}
}
// Post an event to notify that the operation completed.
if (NS_FAILED(mControlEventTarget->Dispatch(
NewRunnableMethod("BackgroundFileSaver::NotifySaveComplete", this,
&BackgroundFileSaver::NotifySaveComplete),
NS_DISPATCH_NORMAL))) {
NS_WARNING("Unable to post completion event to the control thread.");
}
return true;
}
// Called on the control thread.
nsresult BackgroundFileSaver::NotifyTargetChange(nsIFile* aTarget) {
if (mObserver) {
(void)mObserver->OnTargetChange(this, aTarget);
}
return NS_OK;
}
// Called on the control thread.
nsresult BackgroundFileSaver::NotifySaveComplete() {
MOZ_ASSERT(NS_IsMainThread(), "This should be called on the main thread");
nsresult status;
{
MutexAutoLock lock(mLock);
status = mStatus;
}
if (mObserver) {
(void)mObserver->OnSaveComplete(this, status);
// If mObserver keeps alive an enclosure that captures `this`, we'll have a
// cycle that won't be caught by the cycle-collector, so we need to break it
// when we're done here (see bug 1444265).
mObserver = nullptr;
}
// At this point, the worker thread will not process any more events, and we
// can shut it down. Shutting down a thread may re-enter the event loop on
// this thread. This is not a problem in this case, since this function is
// called by a top-level event itself, and we have already invoked the
// completion observer callback. Re-entering the loop can only delay the
// final release and destruction of this saver object, since we are keeping a
// reference to it through the event object.
mBackgroundET = nullptr;
sThreadCount--;
// When there are no more active downloads, we consider the download session
// finished. We record the maximum number of concurrent downloads reached
// during the session in a telemetry histogram, and we reset the maximum
// thread counter for the next download session
if (sThreadCount == 0) {
Telemetry::Accumulate(Telemetry::BACKGROUNDFILESAVER_THREAD_COUNT,
sTelemetryMaxThreadCount);
sTelemetryMaxThreadCount = 0;
}
return NS_OK;
}
nsresult BackgroundFileSaver::ExtractSignatureInfo(const nsAString& filePath) {
MOZ_ASSERT(!NS_IsMainThread(), "Cannot extract signature on main thread");
{
MutexAutoLock lock(mLock);
if (!mSignatureInfoEnabled) {
return NS_OK;
}
}
#ifdef XP_WIN
// Setup the file to check.
WINTRUST_FILE_INFO fileToCheck = {0};
fileToCheck.cbStruct = sizeof(WINTRUST_FILE_INFO);
fileToCheck.pcwszFilePath = filePath.Data();
fileToCheck.hFile = nullptr;
fileToCheck.pgKnownSubject = nullptr;
// We want to check it is signed and trusted.
WINTRUST_DATA trustData = {0};
trustData.cbStruct = sizeof(trustData);
trustData.pPolicyCallbackData = nullptr;
trustData.pSIPClientData = nullptr;
trustData.dwUIChoice = WTD_UI_NONE;
trustData.fdwRevocationChecks = WTD_REVOKE_NONE;
trustData.dwUnionChoice = WTD_CHOICE_FILE;
trustData.dwStateAction = WTD_STATEACTION_VERIFY;
trustData.hWVTStateData = nullptr;
trustData.pwszURLReference = nullptr;
// Disallow revocation checks over the network
trustData.dwProvFlags = WTD_CACHE_ONLY_URL_RETRIEVAL;
// no UI
trustData.dwUIContext = 0;
trustData.pFile = &fileToCheck;
// The WINTRUST_ACTION_GENERIC_VERIFY_V2 policy verifies that the certificate
// chains up to a trusted root CA and has appropriate permissions to sign
// code.
GUID policyGUID = WINTRUST_ACTION_GENERIC_VERIFY_V2;
// Check if the file is signed by something that is trusted. If the file is
// not signed, this is a no-op.
LONG ret = WinVerifyTrust(nullptr, &policyGUID, &trustData);
CRYPT_PROVIDER_DATA* cryptoProviderData = nullptr;
// According to the Windows documentation, we should check against 0 instead
// of ERROR_SUCCESS, which is an HRESULT.
if (ret == 0) {
cryptoProviderData = WTHelperProvDataFromStateData(trustData.hWVTStateData);
}
if (cryptoProviderData) {
// Lock because signature information is read on the main thread.
MutexAutoLock lock(mLock);
LOG(("Downloaded trusted and signed file [this = %p].", this));
// A binary may have multiple signers. Each signer may have multiple certs
// in the chain.
for (DWORD i = 0; i < cryptoProviderData->csSigners; ++i) {
const CERT_CHAIN_CONTEXT* certChainContext =
cryptoProviderData->pasSigners[i].pChainContext;
if (!certChainContext) {
break;
}
for (DWORD j = 0; j < certChainContext->cChain; ++j) {
const CERT_SIMPLE_CHAIN* certSimpleChain =
certChainContext->rgpChain[j];
if (!certSimpleChain) {
break;
}
nsTArray<nsTArray<uint8_t>> certList;
bool extractionSuccess = true;
for (DWORD k = 0; k < certSimpleChain->cElement; ++k) {
CERT_CHAIN_ELEMENT* certChainElement = certSimpleChain->rgpElement[k];
if (certChainElement->pCertContext->dwCertEncodingType !=
X509_ASN_ENCODING) {
continue;
}
nsTArray<uint8_t> cert;
cert.AppendElements(certChainElement->pCertContext->pbCertEncoded,
certChainElement->pCertContext->cbCertEncoded);
certList.AppendElement(std::move(cert));
}
if (extractionSuccess) {
mSignatureInfo.AppendElement(std::move(certList));
}
}
}
// Free the provider data if cryptoProviderData is not null.
trustData.dwStateAction = WTD_STATEACTION_CLOSE;
WinVerifyTrust(nullptr, &policyGUID, &trustData);
} else {
LOG(("Downloaded unsigned or untrusted file [this = %p].", this));
}
#endif
return NS_OK;
}
////////////////////////////////////////////////////////////////////////////////
//// BackgroundFileSaverOutputStream
NS_IMPL_ISUPPORTS(BackgroundFileSaverOutputStream, nsIBackgroundFileSaver,
nsIOutputStream, nsIAsyncOutputStream,
nsIOutputStreamCallback)
BackgroundFileSaverOutputStream::BackgroundFileSaverOutputStream()
: BackgroundFileSaver(), mAsyncWaitCallback(nullptr) {}
bool BackgroundFileSaverOutputStream::HasInfiniteBuffer() { return false; }
nsAsyncCopyProgressFun BackgroundFileSaverOutputStream::GetProgressCallback() {
return nullptr;
}
NS_IMETHODIMP
BackgroundFileSaverOutputStream::Close() { return mPipeOutputStream->Close(); }
NS_IMETHODIMP
BackgroundFileSaverOutputStream::Flush() { return mPipeOutputStream->Flush(); }
NS_IMETHODIMP
BackgroundFileSaverOutputStream::Write(const char* aBuf, uint32_t aCount,
uint32_t* _retval) {
return mPipeOutputStream->Write(aBuf, aCount, _retval);
}
NS_IMETHODIMP
BackgroundFileSaverOutputStream::WriteFrom(nsIInputStream* aFromStream,
uint32_t aCount, uint32_t* _retval) {
return mPipeOutputStream->WriteFrom(aFromStream, aCount, _retval);
}
NS_IMETHODIMP
BackgroundFileSaverOutputStream::WriteSegments(nsReadSegmentFun aReader,
void* aClosure, uint32_t aCount,
uint32_t* _retval) {
return mPipeOutputStream->WriteSegments(aReader, aClosure, aCount, _retval);
}
NS_IMETHODIMP
BackgroundFileSaverOutputStream::IsNonBlocking(bool* _retval) {
return mPipeOutputStream->IsNonBlocking(_retval);
}
NS_IMETHODIMP
BackgroundFileSaverOutputStream::CloseWithStatus(nsresult reason) {
return mPipeOutputStream->CloseWithStatus(reason);
}
NS_IMETHODIMP
BackgroundFileSaverOutputStream::AsyncWait(nsIOutputStreamCallback* aCallback,
uint32_t aFlags,
uint32_t aRequestedCount,
nsIEventTarget* aEventTarget) {
NS_ENSURE_STATE(!mAsyncWaitCallback);
mAsyncWaitCallback = aCallback;
return mPipeOutputStream->AsyncWait(this, aFlags, aRequestedCount,
aEventTarget);
}
NS_IMETHODIMP
BackgroundFileSaverOutputStream::OnOutputStreamReady(
nsIAsyncOutputStream* aStream) {
NS_ENSURE_STATE(mAsyncWaitCallback);
nsCOMPtr<nsIOutputStreamCallback> asyncWaitCallback = nullptr;
asyncWaitCallback.swap(mAsyncWaitCallback);
return asyncWaitCallback->OnOutputStreamReady(this);
}
////////////////////////////////////////////////////////////////////////////////
//// BackgroundFileSaverStreamListener
NS_IMPL_ISUPPORTS(BackgroundFileSaverStreamListener, nsIBackgroundFileSaver,
nsIRequestObserver, nsIStreamListener)
BackgroundFileSaverStreamListener::BackgroundFileSaverStreamListener()
: BackgroundFileSaver(),
mSuspensionLock("BackgroundFileSaverStreamListener.mSuspensionLock"),
mReceivedTooMuchData(false),
mRequest(nullptr),
mRequestSuspended(false) {}
bool BackgroundFileSaverStreamListener::HasInfiniteBuffer() { return true; }
nsAsyncCopyProgressFun
BackgroundFileSaverStreamListener::GetProgressCallback() {
return AsyncCopyProgressCallback;
}
NS_IMETHODIMP
BackgroundFileSaverStreamListener::OnStartRequest(nsIRequest* aRequest) {
NS_ENSURE_ARG(aRequest);
return NS_OK;
}
NS_IMETHODIMP
BackgroundFileSaverStreamListener::OnStopRequest(nsIRequest* aRequest,
nsresult aStatusCode) {
// If an error occurred, cancel the operation immediately. On success, wait
// until the caller has determined whether the file should be renamed.
if (NS_FAILED(aStatusCode)) {
Finish(aStatusCode);
}
return NS_OK;
}
NS_IMETHODIMP
BackgroundFileSaverStreamListener::OnDataAvailable(nsIRequest* aRequest,
nsIInputStream* aInputStream,
uint64_t aOffset,
uint32_t aCount) {
nsresult rv;
NS_ENSURE_ARG(aRequest);
// Read the requested data. Since the pipe has an infinite buffer, we don't
// expect any write error to occur here.
uint32_t writeCount;
rv = mPipeOutputStream->WriteFrom(aInputStream, aCount, &writeCount);
NS_ENSURE_SUCCESS(rv, rv);
// If reading from the input stream fails for any reason, the pipe will return
// a success code, but without reading all the data. Since we should be able
// to read the requested data when OnDataAvailable is called, raise an error.
if (writeCount < aCount) {
NS_WARNING("Reading from the input stream should not have failed.");
return NS_ERROR_UNEXPECTED;
}
bool stateChanged = false;
{
MutexAutoLock lock(mSuspensionLock);
if (!mReceivedTooMuchData) {
uint64_t available;
nsresult rv = mPipeInputStream->Available(&available);
if (NS_SUCCEEDED(rv) && available > REQUEST_SUSPEND_AT) {
mReceivedTooMuchData = true;
mRequest = aRequest;
stateChanged = true;
}
}
}
if (stateChanged) {
NotifySuspendOrResume();
}
return NS_OK;
}
// Called on the worker thread.
// static
void BackgroundFileSaverStreamListener::AsyncCopyProgressCallback(
void* aClosure, uint32_t aCount) {
BackgroundFileSaverStreamListener* self =
(BackgroundFileSaverStreamListener*)aClosure;
// Wait if the control thread is in the process of suspending or resuming.
MutexAutoLock lock(self->mSuspensionLock);
// This function is called when some bytes are consumed by NS_AsyncCopy. Each
// time this happens, verify if a suspended request should be resumed, because
// we have now consumed enough data.
if (self->mReceivedTooMuchData) {
uint64_t available;
nsresult rv = self->mPipeInputStream->Available(&available);
if (NS_FAILED(rv) || available < REQUEST_RESUME_AT) {
self->mReceivedTooMuchData = false;
// Post an event to verify if the request should be resumed.
if (NS_FAILED(self->mControlEventTarget->Dispatch(
NewRunnableMethod(
"BackgroundFileSaverStreamListener::NotifySuspendOrResume",
self,
&BackgroundFileSaverStreamListener::NotifySuspendOrResume),
NS_DISPATCH_NORMAL))) {
NS_WARNING("Unable to post resume event to the control thread.");
}
}
}
}
// Called on the control thread.
nsresult BackgroundFileSaverStreamListener::NotifySuspendOrResume() {
// Prevent the worker thread from changing state while processing.
MutexAutoLock lock(mSuspensionLock);
if (mReceivedTooMuchData) {
if (!mRequestSuspended) {
// Try to suspend the request. If this fails, don't try to resume later.
if (NS_SUCCEEDED(mRequest->Suspend())) {
mRequestSuspended = true;
} else {
NS_WARNING("Unable to suspend the request.");
}
}
} else {
if (mRequestSuspended) {
// Resume the request only if we succeeded in suspending it.
if (NS_SUCCEEDED(mRequest->Resume())) {
mRequestSuspended = false;
} else {
NS_WARNING("Unable to resume the request.");
}
}
}
return NS_OK;
}
////////////////////////////////////////////////////////////////////////////////
//// DigestOutputStream
NS_IMPL_ISUPPORTS(DigestOutputStream, nsIOutputStream)
DigestOutputStream::DigestOutputStream(nsIOutputStream* aStream,
PK11Context* aContext)
: mOutputStream(aStream), mDigestContext(aContext) {
MOZ_ASSERT(mDigestContext, "Can't have null digest context");
MOZ_ASSERT(mOutputStream, "Can't have null output stream");
}
NS_IMETHODIMP
DigestOutputStream::Close() { return mOutputStream->Close(); }
NS_IMETHODIMP
DigestOutputStream::Flush() { return mOutputStream->Flush(); }
NS_IMETHODIMP
DigestOutputStream::Write(const char* aBuf, uint32_t aCount, uint32_t* retval) {
nsresult rv = MapSECStatus(PK11_DigestOp(
mDigestContext, BitwiseCast<const unsigned char*, const char*>(aBuf),
aCount));
NS_ENSURE_SUCCESS(rv, rv);
return mOutputStream->Write(aBuf, aCount, retval);
}
NS_IMETHODIMP
DigestOutputStream::WriteFrom(nsIInputStream* aFromStream, uint32_t aCount,
uint32_t* retval) {
// Not supported. We could read the stream to a buf, call DigestOp on the
// result, seek back and pass the stream on, but it's not worth it since our
// application (NS_AsyncCopy) doesn't invoke this on the sink.
MOZ_CRASH("DigestOutputStream::WriteFrom not implemented");
}
NS_IMETHODIMP
DigestOutputStream::WriteSegments(nsReadSegmentFun aReader, void* aClosure,
uint32_t aCount, uint32_t* retval) {
MOZ_CRASH("DigestOutputStream::WriteSegments not implemented");
}
NS_IMETHODIMP
DigestOutputStream::IsNonBlocking(bool* retval) {
return mOutputStream->IsNonBlocking(retval);
}
#undef LOG_ENABLED
} // namespace net
} // namespace mozilla