blob: 8afc3952f509b4de3693b5aa7a28ac4627f6d6d9 [file] [log] [blame]
//
// Copyright (C) 2012 The Android Open Source Project
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
//
#include "update_engine/update_attempter.h"
#include <stdint.h>
#include <algorithm>
#include <memory>
#include <set>
#include <string>
#include <utility>
#include <vector>
#include <base/bind.h>
#include <base/files/file_util.h>
#include <base/logging.h>
#include <base/rand_util.h>
#include <base/strings/string_util.h>
#include <base/strings/stringprintf.h>
#include <brillo/bind_lambda.h>
#include <brillo/errors/error_codes.h>
#include <brillo/make_unique_ptr.h>
#include <brillo/message_loops/message_loop.h>
#include <policy/device_policy.h>
#include <policy/libpolicy.h>
#include <update_engine/dbus-constants.h>
#include "update_engine/certificate_checker.h"
#include "update_engine/common/boot_control_interface.h"
#include "update_engine/common/clock_interface.h"
#include "update_engine/common/constants.h"
#include "update_engine/common/hardware_interface.h"
#include "update_engine/common/multi_range_http_fetcher.h"
#include "update_engine/common/platform_constants.h"
#include "update_engine/common/prefs_interface.h"
#include "update_engine/common/subprocess.h"
#include "update_engine/common/utils.h"
#include "update_engine/libcurl_http_fetcher.h"
#include "update_engine/metrics.h"
#include "update_engine/omaha_request_action.h"
#include "update_engine/omaha_request_params.h"
#include "update_engine/omaha_response_handler_action.h"
#include "update_engine/p2p_manager.h"
#include "update_engine/payload_consumer/download_action.h"
#include "update_engine/payload_consumer/filesystem_verifier_action.h"
#include "update_engine/payload_consumer/postinstall_runner_action.h"
#include "update_engine/payload_state_interface.h"
#include "update_engine/power_manager_interface.h"
#include "update_engine/system_state.h"
#include "update_engine/update_manager/policy.h"
#include "update_engine/update_manager/update_manager.h"
#include "update_engine/update_status_utils.h"
using base::Bind;
using base::Callback;
using base::Time;
using base::TimeDelta;
using base::TimeTicks;
using brillo::MessageLoop;
using chromeos_update_manager::EvalStatus;
using chromeos_update_manager::Policy;
using chromeos_update_manager::UpdateCheckParams;
using std::set;
using std::shared_ptr;
using std::string;
using std::vector;
namespace chromeos_update_engine {
const int UpdateAttempter::kMaxDeltaUpdateFailures = 3;
namespace {
const int kMaxConsecutiveObeyProxyRequests = 20;
// Minimum threshold to broadcast an status update in progress and time.
const double kBroadcastThresholdProgress = 0.01; // 1%
const int kBroadcastThresholdSeconds = 10;
// By default autest bypasses scattering. If we want to test scattering,
// use kScheduledAUTestURLRequest. The URL used is same in both cases, but
// different params are passed to CheckForUpdate().
const char kAUTestURLRequest[] = "autest";
const char kScheduledAUTestURLRequest[] = "autest-scheduled";
} // namespace
// Turns a generic ErrorCode::kError to a generic error code specific
// to |action| (e.g., ErrorCode::kFilesystemVerifierError). If |code| is
// not ErrorCode::kError, or the action is not matched, returns |code|
// unchanged.
ErrorCode GetErrorCodeForAction(AbstractAction* action,
ErrorCode code) {
if (code != ErrorCode::kError)
return code;
const string type = action->Type();
if (type == OmahaRequestAction::StaticType())
return ErrorCode::kOmahaRequestError;
if (type == OmahaResponseHandlerAction::StaticType())
return ErrorCode::kOmahaResponseHandlerError;
if (type == FilesystemVerifierAction::StaticType())
return ErrorCode::kFilesystemVerifierError;
if (type == PostinstallRunnerAction::StaticType())
return ErrorCode::kPostinstallRunnerError;
return code;
}
UpdateAttempter::UpdateAttempter(SystemState* system_state,
CertificateChecker* cert_checker,
LibCrosProxy* libcros_proxy)
: processor_(new ActionProcessor()),
system_state_(system_state),
#if USE_LIBCROS
cert_checker_(cert_checker),
chrome_proxy_resolver_(libcros_proxy) {
#else
cert_checker_(cert_checker) {
#endif // USE_LIBCROS
}
UpdateAttempter::~UpdateAttempter() {
// CertificateChecker might not be initialized in unittests.
if (cert_checker_)
cert_checker_->SetObserver(nullptr);
// Release ourselves as the ActionProcessor's delegate to prevent
// re-scheduling the updates due to the processing stopped.
processor_->set_delegate(nullptr);
}
void UpdateAttempter::Init() {
// Pulling from the SystemState can only be done after construction, since
// this is an aggregate of various objects (such as the UpdateAttempter),
// which requires them all to be constructed prior to it being used.
prefs_ = system_state_->prefs();
omaha_request_params_ = system_state_->request_params();
if (cert_checker_)
cert_checker_->SetObserver(this);
// In case of update_engine restart without a reboot we need to restore the
// reboot needed state.
if (GetBootTimeAtUpdate(nullptr))
status_ = UpdateStatus::UPDATED_NEED_REBOOT;
else
status_ = UpdateStatus::IDLE;
#if USE_LIBCROS
chrome_proxy_resolver_.Init();
#endif // USE_LIBCROS
}
void UpdateAttempter::ScheduleUpdates() {
if (IsUpdateRunningOrScheduled())
return;
chromeos_update_manager::UpdateManager* const update_manager =
system_state_->update_manager();
CHECK(update_manager);
Callback<void(EvalStatus, const UpdateCheckParams&)> callback = Bind(
&UpdateAttempter::OnUpdateScheduled, base::Unretained(this));
// We limit the async policy request to a reasonably short time, to avoid a
// starvation due to a transient bug.
update_manager->AsyncPolicyRequest(callback, &Policy::UpdateCheckAllowed);
waiting_for_scheduled_check_ = true;
}
void UpdateAttempter::CertificateChecked(ServerToCheck server_to_check,
CertificateCheckResult result) {
metrics::ReportCertificateCheckMetrics(system_state_,
server_to_check,
result);
}
bool UpdateAttempter::CheckAndReportDailyMetrics() {
int64_t stored_value;
Time now = system_state_->clock()->GetWallclockTime();
if (system_state_->prefs()->Exists(kPrefsDailyMetricsLastReportedAt) &&
system_state_->prefs()->GetInt64(kPrefsDailyMetricsLastReportedAt,
&stored_value)) {
Time last_reported_at = Time::FromInternalValue(stored_value);
TimeDelta time_reported_since = now - last_reported_at;
if (time_reported_since.InSeconds() < 0) {
LOG(WARNING) << "Last reported daily metrics "
<< utils::FormatTimeDelta(time_reported_since) << " ago "
<< "which is negative. Either the system clock is wrong or "
<< "the kPrefsDailyMetricsLastReportedAt state variable "
<< "is wrong.";
// In this case, report daily metrics to reset.
} else {
if (time_reported_since.InSeconds() < 24*60*60) {
LOG(INFO) << "Last reported daily metrics "
<< utils::FormatTimeDelta(time_reported_since) << " ago.";
return false;
}
LOG(INFO) << "Last reported daily metrics "
<< utils::FormatTimeDelta(time_reported_since) << " ago, "
<< "which is more than 24 hours ago.";
}
}
LOG(INFO) << "Reporting daily metrics.";
system_state_->prefs()->SetInt64(kPrefsDailyMetricsLastReportedAt,
now.ToInternalValue());
ReportOSAge();
return true;
}
void UpdateAttempter::ReportOSAge() {
struct stat sb;
if (system_state_ == nullptr)
return;
if (stat("/etc/lsb-release", &sb) != 0) {
PLOG(ERROR) << "Error getting file status for /etc/lsb-release "
<< "(Note: this may happen in some unit tests)";
return;
}
Time lsb_release_timestamp = utils::TimeFromStructTimespec(&sb.st_ctim);
Time now = system_state_->clock()->GetWallclockTime();
TimeDelta age = now - lsb_release_timestamp;
if (age.InSeconds() < 0) {
LOG(ERROR) << "The OS age (" << utils::FormatTimeDelta(age)
<< ") is negative. Maybe the clock is wrong? "
<< "(Note: this may happen in some unit tests.)";
return;
}
metrics::ReportDailyMetrics(system_state_, age);
}
void UpdateAttempter::Update(const string& app_version,
const string& omaha_url,
const string& target_channel,
const string& target_version_prefix,
bool obey_proxies,
bool interactive) {
// This is normally called frequently enough so it's appropriate to use as a
// hook for reporting daily metrics.
// TODO(garnold) This should be hooked to a separate (reliable and consistent)
// timeout event.
CheckAndReportDailyMetrics();
// Notify of the new update attempt, clearing prior interactive requests.
if (forced_update_pending_callback_.get())
forced_update_pending_callback_->Run(false, false);
fake_update_success_ = false;
if (status_ == UpdateStatus::UPDATED_NEED_REBOOT) {
// Although we have applied an update, we still want to ping Omaha
// to ensure the number of active statistics is accurate.
//
// Also convey to the UpdateEngine.Check.Result metric that we're
// not performing an update check because of this.
LOG(INFO) << "Not updating b/c we already updated and we're waiting for "
<< "reboot, we'll ping Omaha instead";
metrics::ReportUpdateCheckMetrics(system_state_,
metrics::CheckResult::kRebootPending,
metrics::CheckReaction::kUnset,
metrics::DownloadErrorCode::kUnset);
PingOmaha();
return;
}
if (status_ != UpdateStatus::IDLE) {
// Update in progress. Do nothing
return;
}
if (!CalculateUpdateParams(app_version,
omaha_url,
target_channel,
target_version_prefix,
obey_proxies,
interactive)) {
return;
}
BuildUpdateActions(interactive);
SetStatusAndNotify(UpdateStatus::CHECKING_FOR_UPDATE);
// Update the last check time here; it may be re-updated when an Omaha
// response is received, but this will prevent us from repeatedly scheduling
// checks in the case where a response is not received.
UpdateLastCheckedTime();
// Just in case we didn't update boot flags yet, make sure they're updated
// before any update processing starts.
start_action_processor_ = true;
UpdateBootFlags();
}
void UpdateAttempter::RefreshDevicePolicy() {
// Lazy initialize the policy provider, or reload the latest policy data.
if (!policy_provider_.get())
policy_provider_.reset(new policy::PolicyProvider());
policy_provider_->Reload();
const policy::DevicePolicy* device_policy = nullptr;
if (policy_provider_->device_policy_is_loaded())
device_policy = &policy_provider_->GetDevicePolicy();
if (device_policy)
LOG(INFO) << "Device policies/settings present";
else
LOG(INFO) << "No device policies/settings present.";
system_state_->set_device_policy(device_policy);
system_state_->p2p_manager()->SetDevicePolicy(device_policy);
}
void UpdateAttempter::CalculateP2PParams(bool interactive) {
bool use_p2p_for_downloading = false;
bool use_p2p_for_sharing = false;
// Never use p2p for downloading in interactive checks unless the
// developer has opted in for it via a marker file.
//
// (Why would a developer want to opt in? If he's working on the
// update_engine or p2p codebases so he can actually test his
// code.).
if (system_state_ != nullptr) {
if (!system_state_->p2p_manager()->IsP2PEnabled()) {
LOG(INFO) << "p2p is not enabled - disallowing p2p for both"
<< " downloading and sharing.";
} else {
// Allow p2p for sharing, even in interactive checks.
use_p2p_for_sharing = true;
if (!interactive) {
LOG(INFO) << "Non-interactive check - allowing p2p for downloading";
use_p2p_for_downloading = true;
} else {
LOG(INFO) << "Forcibly disabling use of p2p for downloading "
<< "since this update attempt is interactive.";
}
}
}
PayloadStateInterface* const payload_state = system_state_->payload_state();
payload_state->SetUsingP2PForDownloading(use_p2p_for_downloading);
payload_state->SetUsingP2PForSharing(use_p2p_for_sharing);
}
bool UpdateAttempter::CalculateUpdateParams(const string& app_version,
const string& omaha_url,
const string& target_channel,
const string& target_version_prefix,
bool obey_proxies,
bool interactive) {
http_response_code_ = 0;
PayloadStateInterface* const payload_state = system_state_->payload_state();
// Refresh the policy before computing all the update parameters.
RefreshDevicePolicy();
// Set the target version prefix, if provided.
if (!target_version_prefix.empty())
omaha_request_params_->set_target_version_prefix(target_version_prefix);
CalculateScatteringParams(interactive);
CalculateP2PParams(interactive);
if (payload_state->GetUsingP2PForDownloading() ||
payload_state->GetUsingP2PForSharing()) {
// OK, p2p is to be used - start it and perform housekeeping.
if (!StartP2PAndPerformHousekeeping()) {
// If this fails, disable p2p for this attempt
LOG(INFO) << "Forcibly disabling use of p2p since starting p2p or "
<< "performing housekeeping failed.";
payload_state->SetUsingP2PForDownloading(false);
payload_state->SetUsingP2PForSharing(false);
}
}
if (!omaha_request_params_->Init(app_version,
omaha_url,
interactive)) {
LOG(ERROR) << "Unable to initialize Omaha request params.";
return false;
}
// Set the target channel, if one was provided.
if (target_channel.empty()) {
LOG(INFO) << "No target channel mandated by policy.";
} else {
LOG(INFO) << "Setting target channel as mandated: " << target_channel;
// Pass in false for powerwash_allowed until we add it to the policy
// protobuf.
string error_message;
if (!omaha_request_params_->SetTargetChannel(target_channel, false,
&error_message)) {
LOG(ERROR) << "Setting the channel failed: " << error_message;
}
// Notify observers the target channel change.
BroadcastChannel();
// Since this is the beginning of a new attempt, update the download
// channel. The download channel won't be updated until the next attempt,
// even if target channel changes meanwhile, so that how we'll know if we
// should cancel the current download attempt if there's such a change in
// target channel.
omaha_request_params_->UpdateDownloadChannel();
}
LOG(INFO) << "target_version_prefix = "
<< omaha_request_params_->target_version_prefix()
<< ", scatter_factor_in_seconds = "
<< utils::FormatSecs(scatter_factor_.InSeconds());
LOG(INFO) << "Wall Clock Based Wait Enabled = "
<< omaha_request_params_->wall_clock_based_wait_enabled()
<< ", Update Check Count Wait Enabled = "
<< omaha_request_params_->update_check_count_wait_enabled()
<< ", Waiting Period = " << utils::FormatSecs(
omaha_request_params_->waiting_period().InSeconds());
LOG(INFO) << "Use p2p For Downloading = "
<< payload_state->GetUsingP2PForDownloading()
<< ", Use p2p For Sharing = "
<< payload_state->GetUsingP2PForSharing();
obeying_proxies_ = true;
if (obey_proxies || proxy_manual_checks_ == 0) {
LOG(INFO) << "forced to obey proxies";
// If forced to obey proxies, every 20th request will not use proxies
proxy_manual_checks_++;
LOG(INFO) << "proxy manual checks: " << proxy_manual_checks_;
if (proxy_manual_checks_ >= kMaxConsecutiveObeyProxyRequests) {
proxy_manual_checks_ = 0;
obeying_proxies_ = false;
}
} else if (base::RandInt(0, 4) == 0) {
obeying_proxies_ = false;
}
LOG_IF(INFO, !obeying_proxies_) << "To help ensure updates work, this update "
"check we are ignoring the proxy settings and using "
"direct connections.";
DisableDeltaUpdateIfNeeded();
return true;
}
void UpdateAttempter::CalculateScatteringParams(bool interactive) {
// Take a copy of the old scatter value before we update it, as
// we need to update the waiting period if this value changes.
TimeDelta old_scatter_factor = scatter_factor_;
const policy::DevicePolicy* device_policy = system_state_->device_policy();
if (device_policy) {
int64_t new_scatter_factor_in_secs = 0;
device_policy->GetScatterFactorInSeconds(&new_scatter_factor_in_secs);
if (new_scatter_factor_in_secs < 0) // sanitize input, just in case.
new_scatter_factor_in_secs = 0;
scatter_factor_ = TimeDelta::FromSeconds(new_scatter_factor_in_secs);
}
bool is_scatter_enabled = false;
if (scatter_factor_.InSeconds() == 0) {
LOG(INFO) << "Scattering disabled since scatter factor is set to 0";
} else if (interactive) {
LOG(INFO) << "Scattering disabled as this is an interactive update check";
} else if (system_state_->hardware()->IsOOBEEnabled() &&
!system_state_->hardware()->IsOOBEComplete(nullptr)) {
LOG(INFO) << "Scattering disabled since OOBE is enabled but not complete "
"yet";
} else {
is_scatter_enabled = true;
LOG(INFO) << "Scattering is enabled";
}
if (is_scatter_enabled) {
// This means the scattering policy is turned on.
// Now check if we need to update the waiting period. The two cases
// in which we'd need to update the waiting period are:
// 1. First time in process or a scheduled check after a user-initiated one.
// (omaha_request_params_->waiting_period will be zero in this case).
// 2. Admin has changed the scattering policy value.
// (new scattering value will be different from old one in this case).
int64_t wait_period_in_secs = 0;
if (omaha_request_params_->waiting_period().InSeconds() == 0) {
// First case. Check if we have a suitable value to set for
// the waiting period.
if (prefs_->GetInt64(kPrefsWallClockWaitPeriod, &wait_period_in_secs) &&
wait_period_in_secs > 0 &&
wait_period_in_secs <= scatter_factor_.InSeconds()) {
// This means:
// 1. There's a persisted value for the waiting period available.
// 2. And that persisted value is still valid.
// So, in this case, we should reuse the persisted value instead of
// generating a new random value to improve the chances of a good
// distribution for scattering.
omaha_request_params_->set_waiting_period(
TimeDelta::FromSeconds(wait_period_in_secs));
LOG(INFO) << "Using persisted wall-clock waiting period: " <<
utils::FormatSecs(
omaha_request_params_->waiting_period().InSeconds());
} else {
// This means there's no persisted value for the waiting period
// available or its value is invalid given the new scatter_factor value.
// So, we should go ahead and regenerate a new value for the
// waiting period.
LOG(INFO) << "Persisted value not present or not valid ("
<< utils::FormatSecs(wait_period_in_secs)
<< ") for wall-clock waiting period.";
GenerateNewWaitingPeriod();
}
} else if (scatter_factor_ != old_scatter_factor) {
// This means there's already a waiting period value, but we detected
// a change in the scattering policy value. So, we should regenerate the
// waiting period to make sure it's within the bounds of the new scatter
// factor value.
GenerateNewWaitingPeriod();
} else {
// Neither the first time scattering is enabled nor the scattering value
// changed. Nothing to do.
LOG(INFO) << "Keeping current wall-clock waiting period: " <<
utils::FormatSecs(
omaha_request_params_->waiting_period().InSeconds());
}
// The invariant at this point is that omaha_request_params_->waiting_period
// is non-zero no matter which path we took above.
LOG_IF(ERROR, omaha_request_params_->waiting_period().InSeconds() == 0)
<< "Waiting Period should NOT be zero at this point!!!";
// Since scattering is enabled, wall clock based wait will always be
// enabled.
omaha_request_params_->set_wall_clock_based_wait_enabled(true);
// If we don't have any issues in accessing the file system to update
// the update check count value, we'll turn that on as well.
bool decrement_succeeded = DecrementUpdateCheckCount();
omaha_request_params_->set_update_check_count_wait_enabled(
decrement_succeeded);
} else {
// This means the scattering feature is turned off or disabled for
// this particular update check. Make sure to disable
// all the knobs and artifacts so that we don't invoke any scattering
// related code.
omaha_request_params_->set_wall_clock_based_wait_enabled(false);
omaha_request_params_->set_update_check_count_wait_enabled(false);
omaha_request_params_->set_waiting_period(TimeDelta::FromSeconds(0));
prefs_->Delete(kPrefsWallClockWaitPeriod);
prefs_->Delete(kPrefsUpdateCheckCount);
// Don't delete the UpdateFirstSeenAt file as we don't want manual checks
// that result in no-updates (e.g. due to server side throttling) to
// cause update starvation by having the client generate a new
// UpdateFirstSeenAt for each scheduled check that follows a manual check.
}
}
void UpdateAttempter::GenerateNewWaitingPeriod() {
omaha_request_params_->set_waiting_period(TimeDelta::FromSeconds(
base::RandInt(1, scatter_factor_.InSeconds())));
LOG(INFO) << "Generated new wall-clock waiting period: " << utils::FormatSecs(
omaha_request_params_->waiting_period().InSeconds());
// Do a best-effort to persist this in all cases. Even if the persistence
// fails, we'll still be able to scatter based on our in-memory value.
// The persistence only helps in ensuring a good overall distribution
// across multiple devices if they tend to reboot too often.
system_state_->payload_state()->SetScatteringWaitPeriod(
omaha_request_params_->waiting_period());
}
void UpdateAttempter::BuildPostInstallActions(
InstallPlanAction* previous_action) {
shared_ptr<PostinstallRunnerAction> postinstall_runner_action(
new PostinstallRunnerAction(system_state_->boot_control(),
system_state_->hardware()));
postinstall_runner_action->set_delegate(this);
actions_.push_back(shared_ptr<AbstractAction>(postinstall_runner_action));
BondActions(previous_action,
postinstall_runner_action.get());
}
void UpdateAttempter::BuildUpdateActions(bool interactive) {
CHECK(!processor_->IsRunning());
processor_->set_delegate(this);
// Actions:
std::unique_ptr<LibcurlHttpFetcher> update_check_fetcher(
new LibcurlHttpFetcher(GetProxyResolver(), system_state_->hardware()));
update_check_fetcher->set_server_to_check(ServerToCheck::kUpdate);
// Try harder to connect to the network, esp when not interactive.
// See comment in libcurl_http_fetcher.cc.
update_check_fetcher->set_no_network_max_retries(interactive ? 1 : 3);
shared_ptr<OmahaRequestAction> update_check_action(
new OmahaRequestAction(system_state_,
nullptr,
std::move(update_check_fetcher),
false));
shared_ptr<OmahaResponseHandlerAction> response_handler_action(
new OmahaResponseHandlerAction(system_state_));
shared_ptr<OmahaRequestAction> download_started_action(
new OmahaRequestAction(system_state_,
new OmahaEvent(
OmahaEvent::kTypeUpdateDownloadStarted),
brillo::make_unique_ptr(new LibcurlHttpFetcher(
GetProxyResolver(),
system_state_->hardware())),
false));
LibcurlHttpFetcher* download_fetcher =
new LibcurlHttpFetcher(GetProxyResolver(), system_state_->hardware());
download_fetcher->set_server_to_check(ServerToCheck::kDownload);
shared_ptr<DownloadAction> download_action(new DownloadAction(
prefs_,
system_state_->boot_control(),
system_state_->hardware(),
system_state_,
new MultiRangeHttpFetcher(download_fetcher))); // passes ownership
shared_ptr<OmahaRequestAction> download_finished_action(
new OmahaRequestAction(
system_state_,
new OmahaEvent(OmahaEvent::kTypeUpdateDownloadFinished),
brillo::make_unique_ptr(
new LibcurlHttpFetcher(GetProxyResolver(),
system_state_->hardware())),
false));
shared_ptr<FilesystemVerifierAction> filesystem_verifier_action(
new FilesystemVerifierAction());
shared_ptr<OmahaRequestAction> update_complete_action(
new OmahaRequestAction(
system_state_,
new OmahaEvent(OmahaEvent::kTypeUpdateComplete),
brillo::make_unique_ptr(
new LibcurlHttpFetcher(GetProxyResolver(),
system_state_->hardware())),
false));
download_action->set_delegate(this);
response_handler_action_ = response_handler_action;
download_action_ = download_action;
actions_.push_back(shared_ptr<AbstractAction>(update_check_action));
actions_.push_back(shared_ptr<AbstractAction>(response_handler_action));
actions_.push_back(shared_ptr<AbstractAction>(download_started_action));
actions_.push_back(shared_ptr<AbstractAction>(download_action));
actions_.push_back(shared_ptr<AbstractAction>(download_finished_action));
actions_.push_back(shared_ptr<AbstractAction>(filesystem_verifier_action));
// Bond them together. We have to use the leaf-types when calling
// BondActions().
BondActions(update_check_action.get(),
response_handler_action.get());
BondActions(response_handler_action.get(),
download_action.get());
BondActions(download_action.get(),
filesystem_verifier_action.get());
BuildPostInstallActions(filesystem_verifier_action.get());
actions_.push_back(shared_ptr<AbstractAction>(update_complete_action));
// Enqueue the actions
for (const shared_ptr<AbstractAction>& action : actions_) {
processor_->EnqueueAction(action.get());
}
}
bool UpdateAttempter::Rollback(bool powerwash) {
if (!CanRollback()) {
return false;
}
// Extra check for enterprise-enrolled devices since they don't support
// powerwash.
if (powerwash) {
// Enterprise-enrolled devices have an empty owner in their device policy.
string owner;
RefreshDevicePolicy();
const policy::DevicePolicy* device_policy = system_state_->device_policy();
if (device_policy && (!device_policy->GetOwner(&owner) || owner.empty())) {
LOG(ERROR) << "Enterprise device detected. "
<< "Cannot perform a powerwash for enterprise devices.";
return false;
}
}
processor_->set_delegate(this);
// Initialize the default request params.
if (!omaha_request_params_->Init("", "", true)) {
LOG(ERROR) << "Unable to initialize Omaha request params.";
return false;
}
LOG(INFO) << "Setting rollback options.";
InstallPlan install_plan;
install_plan.target_slot = GetRollbackSlot();
install_plan.source_slot = system_state_->boot_control()->GetCurrentSlot();
TEST_AND_RETURN_FALSE(
install_plan.LoadPartitionsFromSlots(system_state_->boot_control()));
install_plan.powerwash_required = powerwash;
LOG(INFO) << "Using this install plan:";
install_plan.Dump();
shared_ptr<InstallPlanAction> install_plan_action(
new InstallPlanAction(install_plan));
actions_.push_back(shared_ptr<AbstractAction>(install_plan_action));
BuildPostInstallActions(install_plan_action.get());
// Enqueue the actions
for (const shared_ptr<AbstractAction>& action : actions_) {
processor_->EnqueueAction(action.get());
}
// Update the payload state for Rollback.
system_state_->payload_state()->Rollback();
SetStatusAndNotify(UpdateStatus::ATTEMPTING_ROLLBACK);
// Just in case we didn't update boot flags yet, make sure they're updated
// before any update processing starts. This also schedules the start of the
// actions we just posted.
start_action_processor_ = true;
UpdateBootFlags();
return true;
}
bool UpdateAttempter::CanRollback() const {
// We can only rollback if the update_engine isn't busy and we have a valid
// rollback partition.
return (status_ == UpdateStatus::IDLE &&
GetRollbackSlot() != BootControlInterface::kInvalidSlot);
}
BootControlInterface::Slot UpdateAttempter::GetRollbackSlot() const {
LOG(INFO) << "UpdateAttempter::GetRollbackSlot";
const unsigned int num_slots = system_state_->boot_control()->GetNumSlots();
const BootControlInterface::Slot current_slot =
system_state_->boot_control()->GetCurrentSlot();
LOG(INFO) << " Installed slots: " << num_slots;
LOG(INFO) << " Booted from slot: "
<< BootControlInterface::SlotName(current_slot);
if (current_slot == BootControlInterface::kInvalidSlot || num_slots < 2) {
LOG(INFO) << "Device is not updateable.";
return BootControlInterface::kInvalidSlot;
}
vector<BootControlInterface::Slot> bootable_slots;
for (BootControlInterface::Slot slot = 0; slot < num_slots; slot++) {
if (slot != current_slot &&
system_state_->boot_control()->IsSlotBootable(slot)) {
LOG(INFO) << "Found bootable slot "
<< BootControlInterface::SlotName(slot);
return slot;
}
}
LOG(INFO) << "No other bootable slot found.";
return BootControlInterface::kInvalidSlot;
}
void UpdateAttempter::CheckForUpdate(const string& app_version,
const string& omaha_url,
bool interactive) {
LOG(INFO) << "Forced update check requested.";
forced_app_version_.clear();
forced_omaha_url_.clear();
// Certain conditions must be met to allow setting custom version and update
// server URLs. However, kScheduledAUTestURLRequest and kAUTestURLRequest are
// always allowed regardless of device state.
if (IsAnyUpdateSourceAllowed()) {
forced_app_version_ = app_version;
forced_omaha_url_ = omaha_url;
}
if (omaha_url == kScheduledAUTestURLRequest) {
forced_omaha_url_ = constants::kOmahaDefaultAUTestURL;
// Pretend that it's not user-initiated even though it is,
// so as to test scattering logic, etc. which get kicked off
// only in scheduled update checks.
interactive = false;
} else if (omaha_url == kAUTestURLRequest) {
forced_omaha_url_ = constants::kOmahaDefaultAUTestURL;
}
if (forced_update_pending_callback_.get()) {
// Make sure that a scheduling request is made prior to calling the forced
// update pending callback.
ScheduleUpdates();
forced_update_pending_callback_->Run(true, interactive);
}
}
bool UpdateAttempter::RebootIfNeeded() {
if (status_ != UpdateStatus::UPDATED_NEED_REBOOT) {
LOG(INFO) << "Reboot requested, but status is "
<< UpdateStatusToString(status_) << ", so not rebooting.";
return false;
}
if (system_state_->power_manager()->RequestReboot())
return true;
return RebootDirectly();
}
void UpdateAttempter::WriteUpdateCompletedMarker() {
string boot_id;
if (!utils::GetBootId(&boot_id))
return;
prefs_->SetString(kPrefsUpdateCompletedOnBootId, boot_id);
int64_t value = system_state_->clock()->GetBootTime().ToInternalValue();
prefs_->SetInt64(kPrefsUpdateCompletedBootTime, value);
}
bool UpdateAttempter::RebootDirectly() {
vector<string> command;
command.push_back("/sbin/shutdown");
command.push_back("-r");
command.push_back("now");
LOG(INFO) << "Running \"" << base::JoinString(command, " ") << "\"";
int rc = 0;
Subprocess::SynchronousExec(command, &rc, nullptr);
return rc == 0;
}
void UpdateAttempter::OnUpdateScheduled(EvalStatus status,
const UpdateCheckParams& params) {
waiting_for_scheduled_check_ = false;
if (status == EvalStatus::kSucceeded) {
if (!params.updates_enabled) {
LOG(WARNING) << "Updates permanently disabled.";
// Signal disabled status, then switch right back to idle. This is
// necessary for ensuring that observers waiting for a signal change will
// actually notice one on subsequent calls. Note that we don't need to
// re-schedule a check in this case as updates are permanently disabled;
// further (forced) checks may still initiate a scheduling call.
SetStatusAndNotify(UpdateStatus::DISABLED);
SetStatusAndNotify(UpdateStatus::IDLE);
return;
}
LOG(INFO) << "Running "
<< (params.is_interactive ? "interactive" : "periodic")
<< " update.";
Update(forced_app_version_, forced_omaha_url_, params.target_channel,
params.target_version_prefix, false, params.is_interactive);
// Always clear the forced app_version and omaha_url after an update attempt
// so the next update uses the defaults.
forced_app_version_.clear();
forced_omaha_url_.clear();
} else {
LOG(WARNING)
<< "Update check scheduling failed (possibly timed out); retrying.";
ScheduleUpdates();
}
// This check ensures that future update checks will be or are already
// scheduled. The check should never fail. A check failure means that there's
// a bug that will most likely prevent further automatic update checks. It
// seems better to crash in such cases and restart the update_engine daemon
// into, hopefully, a known good state.
CHECK(IsUpdateRunningOrScheduled());
}
void UpdateAttempter::UpdateLastCheckedTime() {
last_checked_time_ = system_state_->clock()->GetWallclockTime().ToTimeT();
}
// Delegate methods:
void UpdateAttempter::ProcessingDone(const ActionProcessor* processor,
ErrorCode code) {
LOG(INFO) << "Processing Done.";
actions_.clear();
// Reset cpu shares back to normal.
cpu_limiter_.StopLimiter();
if (status_ == UpdateStatus::REPORTING_ERROR_EVENT) {
LOG(INFO) << "Error event sent.";
// Inform scheduler of new status;
SetStatusAndNotify(UpdateStatus::IDLE);
ScheduleUpdates();
if (!fake_update_success_) {
return;
}
LOG(INFO) << "Booted from FW B and tried to install new firmware, "
"so requesting reboot from user.";
}
if (code == ErrorCode::kSuccess) {
WriteUpdateCompletedMarker();
prefs_->SetInt64(kPrefsDeltaUpdateFailures, 0);
prefs_->SetString(kPrefsPreviousVersion,
omaha_request_params_->app_version());
DeltaPerformer::ResetUpdateProgress(prefs_, false);
system_state_->payload_state()->UpdateSucceeded();
// Since we're done with scattering fully at this point, this is the
// safest point delete the state files, as we're sure that the status is
// set to reboot (which means no more updates will be applied until reboot)
// This deletion is required for correctness as we want the next update
// check to re-create a new random number for the update check count.
// Similarly, we also delete the wall-clock-wait period that was persisted
// so that we start with a new random value for the next update check
// after reboot so that the same device is not favored or punished in any
// way.
prefs_->Delete(kPrefsUpdateCheckCount);
system_state_->payload_state()->SetScatteringWaitPeriod(TimeDelta());
prefs_->Delete(kPrefsUpdateFirstSeenAt);
SetStatusAndNotify(UpdateStatus::UPDATED_NEED_REBOOT);
ScheduleUpdates();
LOG(INFO) << "Update successfully applied, waiting to reboot.";
// This pointer is null during rollback operations, and the stats
// don't make much sense then anyway.
if (response_handler_action_) {
const InstallPlan& install_plan =
response_handler_action_->install_plan();
// Generate an unique payload identifier.
const string target_version_uid =
install_plan.payload_hash + ":" + install_plan.metadata_signature;
// Expect to reboot into the new version to send the proper metric during
// next boot.
system_state_->payload_state()->ExpectRebootInNewVersion(
target_version_uid);
} else {
// If we just finished a rollback, then we expect to have no Omaha
// response. Otherwise, it's an error.
if (system_state_->payload_state()->GetRollbackVersion().empty()) {
LOG(ERROR) << "Can't send metrics because expected "
"response_handler_action_ missing.";
}
}
return;
}
if (ScheduleErrorEventAction()) {
return;
}
LOG(INFO) << "No update.";
SetStatusAndNotify(UpdateStatus::IDLE);
ScheduleUpdates();
}
void UpdateAttempter::ProcessingStopped(const ActionProcessor* processor) {
// Reset cpu shares back to normal.
cpu_limiter_.StopLimiter();
download_progress_ = 0.0;
SetStatusAndNotify(UpdateStatus::IDLE);
ScheduleUpdates();
actions_.clear();
error_event_.reset(nullptr);
}
// Called whenever an action has finished processing, either successfully
// or otherwise.
void UpdateAttempter::ActionCompleted(ActionProcessor* processor,
AbstractAction* action,
ErrorCode code) {
// Reset download progress regardless of whether or not the download
// action succeeded. Also, get the response code from HTTP request
// actions (update download as well as the initial update check
// actions).
const string type = action->Type();
if (type == DownloadAction::StaticType()) {
download_progress_ = 0.0;
DownloadAction* download_action = static_cast<DownloadAction*>(action);
http_response_code_ = download_action->GetHTTPResponseCode();
} else if (type == OmahaRequestAction::StaticType()) {
OmahaRequestAction* omaha_request_action =
static_cast<OmahaRequestAction*>(action);
// If the request is not an event, then it's the update-check.
if (!omaha_request_action->IsEvent()) {
http_response_code_ = omaha_request_action->GetHTTPResponseCode();
// Record the number of consecutive failed update checks.
if (http_response_code_ == kHttpResponseInternalServerError ||
http_response_code_ == kHttpResponseServiceUnavailable) {
consecutive_failed_update_checks_++;
} else {
consecutive_failed_update_checks_ = 0;
}
// Store the server-dictated poll interval, if any.
server_dictated_poll_interval_ =
std::max(0, omaha_request_action->GetOutputObject().poll_interval);
}
}
if (code != ErrorCode::kSuccess) {
// If the current state is at or past the download phase, count the failure
// in case a switch to full update becomes necessary. Ignore network
// transfer timeouts and failures.
if (status_ >= UpdateStatus::DOWNLOADING &&
code != ErrorCode::kDownloadTransferError) {
MarkDeltaUpdateFailure();
}
// On failure, schedule an error event to be sent to Omaha.
CreatePendingErrorEvent(action, code);
return;
}
// Find out which action completed.
if (type == OmahaResponseHandlerAction::StaticType()) {
// Note that the status will be updated to DOWNLOADING when some bytes get
// actually downloaded from the server and the BytesReceived callback is
// invoked. This avoids notifying the user that a download has started in
// cases when the server and the client are unable to initiate the download.
CHECK(action == response_handler_action_.get());
const InstallPlan& plan = response_handler_action_->install_plan();
UpdateLastCheckedTime();
new_version_ = plan.version;
new_payload_size_ = plan.payload_size;
SetupDownload();
cpu_limiter_.StartLimiter();
SetStatusAndNotify(UpdateStatus::UPDATE_AVAILABLE);
} else if (type == DownloadAction::StaticType()) {
SetStatusAndNotify(UpdateStatus::FINALIZING);
}
}
void UpdateAttempter::BytesReceived(uint64_t bytes_progressed,
uint64_t bytes_received,
uint64_t total) {
// The PayloadState keeps track of how many bytes were actually downloaded
// from a given URL for the URL skipping logic.
system_state_->payload_state()->DownloadProgress(bytes_progressed);
double progress = 0;
if (total)
progress = static_cast<double>(bytes_received) / static_cast<double>(total);
if (status_ != UpdateStatus::DOWNLOADING || bytes_received == total) {
download_progress_ = progress;
SetStatusAndNotify(UpdateStatus::DOWNLOADING);
} else {
ProgressUpdate(progress);
}
}
void UpdateAttempter::DownloadComplete() {
system_state_->payload_state()->DownloadComplete();
}
bool UpdateAttempter::OnCheckForUpdates(brillo::ErrorPtr* error) {
CheckForUpdate(
"" /* app_version */, "" /* omaha_url */, true /* interactive */);
return true;
}
bool UpdateAttempter::OnTrackChannel(const string& channel,
brillo::ErrorPtr* error) {
LOG(INFO) << "Setting destination channel to: " << channel;
string error_message;
if (!system_state_->request_params()->SetTargetChannel(
channel, false /* powerwash_allowed */, &error_message)) {
brillo::Error::AddTo(error,
FROM_HERE,
brillo::errors::dbus::kDomain,
"set_target_error",
error_message);
return false;
}
// Notify observers the target channel change.
BroadcastChannel();
return true;
}
bool UpdateAttempter::GetWeaveState(int64_t* last_checked_time,
double* progress,
UpdateStatus* update_status,
string* current_channel,
string* tracking_channel) {
*last_checked_time = last_checked_time_;
*progress = download_progress_;
*update_status = status_;
OmahaRequestParams* rp = system_state_->request_params();
*current_channel = rp->current_channel();
*tracking_channel = rp->target_channel();
return true;
}
void UpdateAttempter::ProgressUpdate(double progress) {
// Self throttle based on progress. Also send notifications if progress is
// too slow.
if (progress == 1.0 ||
progress - download_progress_ >= kBroadcastThresholdProgress ||
TimeTicks::Now() - last_notify_time_ >=
TimeDelta::FromSeconds(kBroadcastThresholdSeconds)) {
download_progress_ = progress;
BroadcastStatus();
}
}
bool UpdateAttempter::ResetStatus() {
LOG(INFO) << "Attempting to reset state from "
<< UpdateStatusToString(status_) << " to UpdateStatus::IDLE";
switch (status_) {
case UpdateStatus::IDLE:
// no-op.
return true;
case UpdateStatus::UPDATED_NEED_REBOOT: {
bool ret_value = true;
status_ = UpdateStatus::IDLE;
// Remove the reboot marker so that if the machine is rebooted
// after resetting to idle state, it doesn't go back to
// UpdateStatus::UPDATED_NEED_REBOOT state.
ret_value = prefs_->Delete(kPrefsUpdateCompletedOnBootId) && ret_value;
ret_value = prefs_->Delete(kPrefsUpdateCompletedBootTime) && ret_value;
// Update the boot flags so the current slot has higher priority.
BootControlInterface* boot_control = system_state_->boot_control();
if (!boot_control->SetActiveBootSlot(boot_control->GetCurrentSlot()))
ret_value = false;
// Notify the PayloadState that the successful payload was canceled.
system_state_->payload_state()->ResetUpdateStatus();
// The previous version is used to report back to omaha after reboot that
// we actually rebooted into the new version from this "prev-version". We
// need to clear out this value now to prevent it being sent on the next
// updatecheck request.
ret_value = prefs_->SetString(kPrefsPreviousVersion, "") && ret_value;
LOG(INFO) << "Reset status " << (ret_value ? "successful" : "failed");
return ret_value;
}
default:
LOG(ERROR) << "Reset not allowed in this state.";
return false;
}
}
bool UpdateAttempter::GetStatus(int64_t* last_checked_time,
double* progress,
string* current_operation,
string* new_version,
int64_t* new_payload_size) {
*last_checked_time = last_checked_time_;
*progress = download_progress_;
*current_operation = UpdateStatusToString(status_);
*new_version = new_version_;
*new_payload_size = new_payload_size_;
return true;
}
void UpdateAttempter::UpdateBootFlags() {
if (update_boot_flags_running_) {
LOG(INFO) << "Update boot flags running, nothing to do.";
return;
}
if (updated_boot_flags_) {
LOG(INFO) << "Already updated boot flags. Skipping.";
if (start_action_processor_) {
ScheduleProcessingStart();
}
return;
}
// This is purely best effort. Failures should be logged by Subprocess. Run
// the script asynchronously to avoid blocking the event loop regardless of
// the script runtime.
update_boot_flags_running_ = true;
LOG(INFO) << "Marking booted slot as good.";
if (!system_state_->boot_control()->MarkBootSuccessfulAsync(Bind(
&UpdateAttempter::CompleteUpdateBootFlags, base::Unretained(this)))) {
LOG(ERROR) << "Failed to mark current boot as successful.";
CompleteUpdateBootFlags(false);
}
}
void UpdateAttempter::CompleteUpdateBootFlags(bool successful) {
update_boot_flags_running_ = false;
updated_boot_flags_ = true;
if (start_action_processor_) {
ScheduleProcessingStart();
}
}
void UpdateAttempter::BroadcastStatus() {
for (const auto& observer : service_observers_) {
observer->SendStatusUpdate(last_checked_time_,
download_progress_,
status_,
new_version_,
new_payload_size_);
}
last_notify_time_ = TimeTicks::Now();
}
void UpdateAttempter::BroadcastChannel() {
for (const auto& observer : service_observers_) {
observer->SendChannelChangeUpdate(
system_state_->request_params()->target_channel());
}
}
uint32_t UpdateAttempter::GetErrorCodeFlags() {
uint32_t flags = 0;
if (!system_state_->hardware()->IsNormalBootMode())
flags |= static_cast<uint32_t>(ErrorCode::kDevModeFlag);
if (response_handler_action_.get() &&
response_handler_action_->install_plan().is_resume)
flags |= static_cast<uint32_t>(ErrorCode::kResumedFlag);
if (!system_state_->hardware()->IsOfficialBuild())
flags |= static_cast<uint32_t>(ErrorCode::kTestImageFlag);
if (omaha_request_params_->update_url() !=
constants::kOmahaDefaultProductionURL) {
flags |= static_cast<uint32_t>(ErrorCode::kTestOmahaUrlFlag);
}
return flags;
}
bool UpdateAttempter::ShouldCancel(ErrorCode* cancel_reason) {
// Check if the channel we're attempting to update to is the same as the
// target channel currently chosen by the user.
OmahaRequestParams* params = system_state_->request_params();
if (params->download_channel() != params->target_channel()) {
LOG(ERROR) << "Aborting download as target channel: "
<< params->target_channel()
<< " is different from the download channel: "
<< params->download_channel();
*cancel_reason = ErrorCode::kUpdateCanceledByChannelChange;
return true;
}
return false;
}
void UpdateAttempter::SetStatusAndNotify(UpdateStatus status) {
status_ = status;
BroadcastStatus();
}
void UpdateAttempter::CreatePendingErrorEvent(AbstractAction* action,
ErrorCode code) {
if (error_event_.get()) {
// This shouldn't really happen.
LOG(WARNING) << "There's already an existing pending error event.";
return;
}
// For now assume that a generic Omaha response action failure means that
// there's no update so don't send an event. Also, double check that the
// failure has not occurred while sending an error event -- in which case
// don't schedule another. This shouldn't really happen but just in case...
if ((action->Type() == OmahaResponseHandlerAction::StaticType() &&
code == ErrorCode::kError) ||
status_ == UpdateStatus::REPORTING_ERROR_EVENT) {
return;
}
// Classify the code to generate the appropriate result so that
// the Borgmon charts show up the results correctly.
// Do this before calling GetErrorCodeForAction which could potentially
// augment the bit representation of code and thus cause no matches for
// the switch cases below.
OmahaEvent::Result event_result;
switch (code) {
case ErrorCode::kOmahaUpdateIgnoredPerPolicy:
case ErrorCode::kOmahaUpdateDeferredPerPolicy:
case ErrorCode::kOmahaUpdateDeferredForBackoff:
event_result = OmahaEvent::kResultUpdateDeferred;
break;
default:
event_result = OmahaEvent::kResultError;
break;
}
code = GetErrorCodeForAction(action, code);
fake_update_success_ = code == ErrorCode::kPostinstallBootedFromFirmwareB;
// Compute the final error code with all the bit flags to be sent to Omaha.
code = static_cast<ErrorCode>(
static_cast<uint32_t>(code) | GetErrorCodeFlags());
error_event_.reset(new OmahaEvent(OmahaEvent::kTypeUpdateComplete,
event_result,
code));
}
bool UpdateAttempter::ScheduleErrorEventAction() {
if (error_event_.get() == nullptr)
return false;
LOG(ERROR) << "Update failed.";
system_state_->payload_state()->UpdateFailed(error_event_->error_code);
// Send it to Omaha.
LOG(INFO) << "Reporting the error event";
shared_ptr<OmahaRequestAction> error_event_action(
new OmahaRequestAction(system_state_,
error_event_.release(), // Pass ownership.
brillo::make_unique_ptr(new LibcurlHttpFetcher(
GetProxyResolver(),
system_state_->hardware())),
false));
actions_.push_back(shared_ptr<AbstractAction>(error_event_action));
processor_->EnqueueAction(error_event_action.get());
SetStatusAndNotify(UpdateStatus::REPORTING_ERROR_EVENT);
processor_->StartProcessing();
return true;
}
void UpdateAttempter::ScheduleProcessingStart() {
LOG(INFO) << "Scheduling an action processor start.";
start_action_processor_ = false;
MessageLoop::current()->PostTask(
FROM_HERE,
Bind([](ActionProcessor* processor) { processor->StartProcessing(); },
base::Unretained(processor_.get())));
}
void UpdateAttempter::DisableDeltaUpdateIfNeeded() {
int64_t delta_failures;
if (omaha_request_params_->delta_okay() &&
prefs_->GetInt64(kPrefsDeltaUpdateFailures, &delta_failures) &&
delta_failures >= kMaxDeltaUpdateFailures) {
LOG(WARNING) << "Too many delta update failures, forcing full update.";
omaha_request_params_->set_delta_okay(false);
}
}
void UpdateAttempter::MarkDeltaUpdateFailure() {
// Don't try to resume a failed delta update.
DeltaPerformer::ResetUpdateProgress(prefs_, false);
int64_t delta_failures;
if (!prefs_->GetInt64(kPrefsDeltaUpdateFailures, &delta_failures) ||
delta_failures < 0) {
delta_failures = 0;
}
prefs_->SetInt64(kPrefsDeltaUpdateFailures, ++delta_failures);
}
void UpdateAttempter::SetupDownload() {
MultiRangeHttpFetcher* fetcher =
static_cast<MultiRangeHttpFetcher*>(download_action_->http_fetcher());
fetcher->ClearRanges();
if (response_handler_action_->install_plan().is_resume) {
// Resuming an update so fetch the update manifest metadata first.
int64_t manifest_metadata_size = 0;
int64_t manifest_signature_size = 0;
prefs_->GetInt64(kPrefsManifestMetadataSize, &manifest_metadata_size);
prefs_->GetInt64(kPrefsManifestSignatureSize, &manifest_signature_size);
fetcher->AddRange(0, manifest_metadata_size + manifest_signature_size);
// If there're remaining unprocessed data blobs, fetch them. Be careful not
// to request data beyond the end of the payload to avoid 416 HTTP response
// error codes.
int64_t next_data_offset = 0;
prefs_->GetInt64(kPrefsUpdateStateNextDataOffset, &next_data_offset);
uint64_t resume_offset =
manifest_metadata_size + manifest_signature_size + next_data_offset;
if (resume_offset < response_handler_action_->install_plan().payload_size) {
fetcher->AddRange(resume_offset);
}
} else {
fetcher->AddRange(0);
}
}
void UpdateAttempter::PingOmaha() {
if (!processor_->IsRunning()) {
shared_ptr<OmahaRequestAction> ping_action(new OmahaRequestAction(
system_state_,
nullptr,
brillo::make_unique_ptr(new LibcurlHttpFetcher(
GetProxyResolver(),
system_state_->hardware())),
true));
actions_.push_back(shared_ptr<OmahaRequestAction>(ping_action));
processor_->set_delegate(nullptr);
processor_->EnqueueAction(ping_action.get());
// Call StartProcessing() synchronously here to avoid any race conditions
// caused by multiple outstanding ping Omaha requests. If we call
// StartProcessing() asynchronously, the device can be suspended before we
// get a chance to callback to StartProcessing(). When the device resumes
// (assuming the device sleeps longer than the next update check period),
// StartProcessing() is called back and at the same time, the next update
// check is fired which eventually invokes StartProcessing(). A crash
// can occur because StartProcessing() checks to make sure that the
// processor is idle which it isn't due to the two concurrent ping Omaha
// requests.
processor_->StartProcessing();
} else {
LOG(WARNING) << "Action processor running, Omaha ping suppressed.";
}
// Update the last check time here; it may be re-updated when an Omaha
// response is received, but this will prevent us from repeatedly scheduling
// checks in the case where a response is not received.
UpdateLastCheckedTime();
// Update the status which will schedule the next update check
SetStatusAndNotify(UpdateStatus::UPDATED_NEED_REBOOT);
ScheduleUpdates();
}
bool UpdateAttempter::DecrementUpdateCheckCount() {
int64_t update_check_count_value;
if (!prefs_->Exists(kPrefsUpdateCheckCount)) {
// This file does not exist. This means we haven't started our update
// check count down yet, so nothing more to do. This file will be created
// later when we first satisfy the wall-clock-based-wait period.
LOG(INFO) << "No existing update check count. That's normal.";
return true;
}
if (prefs_->GetInt64(kPrefsUpdateCheckCount, &update_check_count_value)) {
// Only if we're able to read a proper integer value, then go ahead
// and decrement and write back the result in the same file, if needed.
LOG(INFO) << "Update check count = " << update_check_count_value;
if (update_check_count_value == 0) {
// It could be 0, if, for some reason, the file didn't get deleted
// when we set our status to waiting for reboot. so we just leave it
// as is so that we can prevent another update_check wait for this client.
LOG(INFO) << "Not decrementing update check count as it's already 0.";
return true;
}
if (update_check_count_value > 0)
update_check_count_value--;
else
update_check_count_value = 0;
// Write out the new value of update_check_count_value.
if (prefs_->SetInt64(kPrefsUpdateCheckCount, update_check_count_value)) {
// We successfully wrote out te new value, so enable the
// update check based wait.
LOG(INFO) << "New update check count = " << update_check_count_value;
return true;
}
}
LOG(INFO) << "Deleting update check count state due to read/write errors.";
// We cannot read/write to the file, so disable the update check based wait
// so that we don't get stuck in this OS version by any chance (which could
// happen if there's some bug that causes to read/write incorrectly).
// Also attempt to delete the file to do our best effort to cleanup.
prefs_->Delete(kPrefsUpdateCheckCount);
return false;
}
void UpdateAttempter::UpdateEngineStarted() {
// If we just booted into a new update, keep the previous OS version
// in case we rebooted because of a crash of the old version, so we
// can do a proper crash report with correct information.
// This must be done before calling
// system_state_->payload_state()->UpdateEngineStarted() since it will
// delete SystemUpdated marker file.
if (system_state_->system_rebooted() &&
prefs_->Exists(kPrefsSystemUpdatedMarker)) {
if (!prefs_->GetString(kPrefsPreviousVersion, &prev_version_)) {
// If we fail to get the version string, make sure it stays empty.
prev_version_.clear();
}
}
system_state_->payload_state()->UpdateEngineStarted();
StartP2PAtStartup();
}
bool UpdateAttempter::StartP2PAtStartup() {
if (system_state_ == nullptr ||
!system_state_->p2p_manager()->IsP2PEnabled()) {
LOG(INFO) << "Not starting p2p at startup since it's not enabled.";
return false;
}
if (system_state_->p2p_manager()->CountSharedFiles() < 1) {
LOG(INFO) << "Not starting p2p at startup since our application "
<< "is not sharing any files.";
return false;
}
return StartP2PAndPerformHousekeeping();
}
bool UpdateAttempter::StartP2PAndPerformHousekeeping() {
if (system_state_ == nullptr)
return false;
if (!system_state_->p2p_manager()->IsP2PEnabled()) {
LOG(INFO) << "Not starting p2p since it's not enabled.";
return false;
}
LOG(INFO) << "Ensuring that p2p is running.";
if (!system_state_->p2p_manager()->EnsureP2PRunning()) {
LOG(ERROR) << "Error starting p2p.";
return false;
}
LOG(INFO) << "Performing p2p housekeeping.";
if (!system_state_->p2p_manager()->PerformHousekeeping()) {
LOG(ERROR) << "Error performing housekeeping for p2p.";
return false;
}
LOG(INFO) << "Done performing p2p housekeeping.";
return true;
}
bool UpdateAttempter::GetBootTimeAtUpdate(Time *out_boot_time) {
// In case of an update_engine restart without a reboot, we stored the boot_id
// when the update was completed by setting a pref, so we can check whether
// the last update was on this boot or a previous one.
string boot_id;
TEST_AND_RETURN_FALSE(utils::GetBootId(&boot_id));
string update_completed_on_boot_id;
if (!prefs_->Exists(kPrefsUpdateCompletedOnBootId) ||
!prefs_->GetString(kPrefsUpdateCompletedOnBootId,
&update_completed_on_boot_id) ||
update_completed_on_boot_id != boot_id)
return false;
// Short-circuit avoiding the read in case out_boot_time is nullptr.
if (out_boot_time) {
int64_t boot_time = 0;
// Since the kPrefsUpdateCompletedOnBootId was correctly set, this pref
// should not fail.
TEST_AND_RETURN_FALSE(
prefs_->GetInt64(kPrefsUpdateCompletedBootTime, &boot_time));
*out_boot_time = Time::FromInternalValue(boot_time);
}
return true;
}
bool UpdateAttempter::IsUpdateRunningOrScheduled() {
return ((status_ != UpdateStatus::IDLE &&
status_ != UpdateStatus::UPDATED_NEED_REBOOT) ||
waiting_for_scheduled_check_);
}
bool UpdateAttempter::IsAnyUpdateSourceAllowed() {
// We allow updates from any source if either of these are true:
// * The device is running an unofficial (dev/test) image.
// * The debugd dev features are accessible (i.e. in devmode with no owner).
// This protects users running a base image, while still allowing a specific
// window (gated by the debug dev features) where `cros flash` is usable.
if (!system_state_->hardware()->IsOfficialBuild()) {
LOG(INFO) << "Non-official build; allowing any update source.";
return true;
}
if (system_state_->hardware()->AreDevFeaturesEnabled()) {
LOG(INFO) << "Developer features enabled; allowing custom update sources.";
return true;
}
LOG(INFO)
<< "Developer features disabled; disallowing custom update sources.";
return false;
}
} // namespace chromeos_update_engine