| // Copyright 2018 The Fuchsia Authors. All rights reserved. |
| // Use of this source code is governed by a BSD-style license that can be |
| // found in the LICENSE file. |
| |
| #include "src/developer/debug/debug_agent/debugged_process.h" |
| |
| #include <inttypes.h> |
| #include <lib/fit/defer.h> |
| #include <lib/syslog/cpp/macros.h> |
| #include <lib/zx/clock.h> |
| #include <lib/zx/time.h> |
| #include <zircon/syscalls/exception.h> |
| |
| #include <utility> |
| |
| #include "src/developer/debug/debug_agent/align.h" |
| #include "src/developer/debug/debug_agent/debug_agent.h" |
| #include "src/developer/debug/debug_agent/debugged_thread.h" |
| #include "src/developer/debug/debug_agent/elf_utils.h" |
| #include "src/developer/debug/debug_agent/exception_handle.h" |
| #include "src/developer/debug/debug_agent/hardware_breakpoint.h" |
| #include "src/developer/debug/debug_agent/process_breakpoint.h" |
| #include "src/developer/debug/debug_agent/software_breakpoint.h" |
| #include "src/developer/debug/debug_agent/thread_handle.h" |
| #include "src/developer/debug/debug_agent/watchpoint.h" |
| #include "src/developer/debug/ipc/agent_protocol.h" |
| #include "src/developer/debug/ipc/message_reader.h" |
| #include "src/developer/debug/ipc/message_writer.h" |
| #include "src/developer/debug/shared/logging/logging.h" |
| #include "src/developer/debug/shared/platform_message_loop.h" |
| #include "src/developer/debug/shared/zx_status.h" |
| #include "src/lib/fxl/strings/string_printf.h" |
| |
| namespace debug_agent { |
| |
| namespace { |
| |
| std::vector<char> ReadSocketInput(debug_ipc::BufferedZxSocket* socket) { |
| FX_DCHECK(socket->valid()); |
| |
| constexpr size_t kReadSize = 1024; // Read in 1K chunks. |
| |
| std::vector<char> data; |
| auto& stream = socket->stream(); |
| while (true) { |
| char buf[kReadSize]; |
| |
| size_t read_amount = stream.Read(buf, kReadSize); |
| data.insert(data.end(), buf, buf + read_amount); |
| |
| if (read_amount < kReadSize) |
| break; |
| } |
| return data; |
| } |
| |
| // Meant to be used in debug logging. |
| std::string LogPreamble(const DebuggedProcess* process) { |
| return fxl::StringPrintf("[P: %lu (%s)] ", process->koid(), |
| process->process_handle().GetName().c_str()); |
| } |
| |
| void LogRegisterBreakpoint(debug_ipc::FileLineFunction location, DebuggedProcess* process, |
| Breakpoint* bp, uint64_t address) { |
| if (!debug_ipc::IsDebugModeActive()) |
| return; |
| |
| std::stringstream ss; |
| ss << LogPreamble(process) << "Setting breakpoint " << bp->settings().id << " (" |
| << bp->settings().name << ") on 0x" << std::hex << address; |
| |
| if (bp->settings().one_shot) |
| ss << " (one shot)"; |
| |
| DEBUG_LOG_WITH_LOCATION(Process, location) << ss.str(); |
| } |
| |
| } // namespace |
| |
| // DebuggedProcessCreateInfo ----------------------------------------------------------------------- |
| |
| DebuggedProcessCreateInfo::DebuggedProcessCreateInfo(std::unique_ptr<ProcessHandle> handle) |
| : handle(std::move(handle)) {} |
| |
| // DebuggedProcess --------------------------------------------------------------------------------- |
| |
| DebuggedProcess::DebuggedProcess(DebugAgent* debug_agent, DebuggedProcessCreateInfo&& create_info) |
| : debug_agent_(debug_agent), |
| process_handle_(std::move(create_info.handle)), |
| from_limbo_(create_info.from_limbo) { |
| // If create_info out or err are not valid, calling Init on the |
| // BufferedZxSocket will fail and leave it in an invalid state. This is |
| // expected if the io sockets could be obtained from the inferior. |
| stdout_.Init(std::move(create_info.out)); |
| stderr_.Init(std::move(create_info.err)); |
| } |
| |
| DebuggedProcess::~DebuggedProcess() { DetachFromProcess(); } |
| |
| void DebuggedProcess::DetachFromProcess() { |
| // 1. Remove installed software breakpoints. |
| // We need to tell each thread that this will happen. |
| for (auto& [address, breakpoint] : software_breakpoints_) { |
| for (auto& [thread_koid, thread] : threads_) { |
| thread->WillDeleteProcessBreakpoint(breakpoint.get()); |
| } |
| } |
| |
| // Clear the resources. |
| software_breakpoints_.clear(); |
| hardware_breakpoints_.clear(); |
| watchpoints_.clear(); |
| |
| // 2. Resume threads. |
| // Technically a 0'ed request would work, but being explicit is future-proof. |
| debug_ipc::ResumeRequest resume_request = {}; |
| resume_request.how = debug_ipc::ResumeRequest::How::kResolveAndContinue; |
| resume_request.process_koid = koid(); |
| OnResume(resume_request); |
| |
| // 3. Unbind from notifications (this will detach from the process). |
| process_handle_->Detach(); |
| } |
| |
| zx_status_t DebuggedProcess::Init() { |
| // Watch for process events. |
| if (zx_status_t status = process_handle_->Attach(this); status != ZX_OK) |
| return status; |
| |
| RegisterDebugState(); |
| |
| // Binding stdout/stderr. |
| // We bind |this| into the callbacks. This is OK because the DebuggedProcess |
| // owns both sockets, meaning that it's assured to outlive the sockets. |
| |
| if (stdout_.valid()) { |
| stdout_.set_data_available_callback([this]() { OnStdout(false); }); |
| stdout_.set_error_callback([this]() { OnStdout(true); }); |
| if (zx_status_t status = stdout_.Start(); status != ZX_OK) { |
| FX_LOGS(WARNING) << "Could not listen on stdout for process " << process_handle_->GetName() |
| << ": " << debug_ipc::ZxStatusToString(status); |
| stdout_.Reset(); |
| } |
| } |
| |
| if (stderr_.valid()) { |
| stderr_.set_data_available_callback([this]() { OnStderr(false); }); |
| stderr_.set_error_callback([this]() { OnStderr(true); }); |
| if (zx_status_t status = stderr_.Start(); status != ZX_OK) { |
| FX_LOGS(WARNING) << "Could not listen on stderr for process " << process_handle_->GetName() |
| << ": " << debug_ipc::ZxStatusToString(status); |
| stderr_.Reset(); |
| } |
| } |
| |
| return ZX_OK; |
| } |
| |
| void DebuggedProcess::OnResume(const debug_ipc::ResumeRequest& request) { |
| if (request.thread_koids.empty()) { |
| // Empty thread ID list means resume all threads. |
| for (auto& [thread_koid, thread] : threads_) |
| thread->ClientResume(request); |
| } else { |
| for (uint64_t thread_koid : request.thread_koids) { |
| if (DebuggedThread* thread = GetThread(thread_koid)) |
| thread->ClientResume(request); |
| // Might be not found if there is a race between the thread exiting and the client sending the |
| // request. |
| } |
| } |
| } |
| |
| void DebuggedProcess::OnReadMemory(const debug_ipc::ReadMemoryRequest& request, |
| debug_ipc::ReadMemoryReply* reply) { |
| reply->blocks = process_handle_->ReadMemoryBlocks(request.address, request.size); |
| |
| // Remove any breakpoint instructions we've inserted. |
| // |
| // If there are a lot of ProcessBreakpoints this will get slow. If we find we have 100's of |
| // breakpoints an auxiliary data structure could be added to find overlapping breakpoints faster. |
| for (const auto& [addr, bp] : software_breakpoints_) { |
| // Generally there will be only one block. If we start reading many megabytes that cross |
| // mapped memory boundaries, a top-level range check would be a good idea to avoid unnecessary |
| // iteration. |
| for (auto& block : reply->blocks) { |
| bp->FixupMemoryBlock(&block); |
| } |
| } |
| } |
| |
| void DebuggedProcess::OnKill(const debug_ipc::KillRequest& request, debug_ipc::KillReply* reply) { |
| // Stop observing before killing the process to avoid getting exceptions after we stopped |
| // listening to them. |
| process_handle_->Detach(); |
| |
| // Since we're being killed, we treat this process as not having any more |
| // threads. This makes cleanup code more straightforward, as there are no |
| // threads to resume/handle. |
| threads_.clear(); |
| reply->status = process_handle_->Kill(); |
| } |
| |
| DebuggedThread* DebuggedProcess::GetThread(zx_koid_t thread_koid) const { |
| auto found_thread = threads_.find(thread_koid); |
| if (found_thread == threads_.end()) |
| return nullptr; |
| return found_thread->second.get(); |
| } |
| |
| std::vector<DebuggedThread*> DebuggedProcess::GetThreads() const { |
| std::vector<DebuggedThread*> threads; |
| threads.reserve(threads_.size()); |
| for (auto& kv : threads_) |
| threads.emplace_back(kv.second.get()); |
| return threads; |
| } |
| |
| void DebuggedProcess::PopulateCurrentThreads() { |
| for (auto& thread : process_handle_->GetChildThreads()) { |
| // We should never populate the same thread twice. |
| zx_koid_t thread_koid = thread->GetKoid(); |
| if (threads_.find(thread_koid) != threads_.end()) |
| continue; |
| |
| auto new_thread = std::make_unique<DebuggedThread>(debug_agent_, this, std::move(thread), |
| ThreadCreationOption::kRunningKeepRunning); |
| threads_.emplace(thread_koid, std::move(new_thread)); |
| } |
| } |
| |
| std::vector<debug_ipc::ThreadRecord> DebuggedProcess::GetThreadRecords() const { |
| std::vector<debug_ipc::ThreadRecord> result; |
| for (const auto& pair : threads_) |
| result.push_back(pair.second->GetThreadRecord(debug_ipc::ThreadRecord::StackAmount::kMinimal)); |
| return result; |
| } |
| |
| bool DebuggedProcess::RegisterDebugState() { |
| // HOW REGISTRATION WITH THE LOADER WORKS. |
| // |
| // Upon process initialization and before executing the normal program code, ld.so sets the |
| // ZX_PROP_PROCESS_DEBUG_ADDR property on its own process to the address of a known struct defined |
| // in <link.h> containing the state of the loader. Debuggers can come along later, get the address |
| // from this property, and inspect the state of the dynamic loader for this process (get the |
| // loaded libraries, set breakpoints for loads, etc.). |
| // |
| // When launching a process in a debugger, the debugger needs to know when this property has been |
| // set or there will be a race to know when it's valid. To resolve this, the debuggers sets a |
| // known magic value to the property before startup. The loader checks for this value when setting |
| // the property, and if it had the magic value, issues a hardcoded software breakpoint. The |
| // debugger catches this breakpoint exception, reads the now-valid address from the property, and |
| // continues initialization. |
| // |
| // It's also possible that the property has been properly set up prior to starting the process. |
| // In Posix this can happen with a fork() where the entire process is duplicated, including the |
| // loader state and all dynamically loaded libraries. In Zircon this can happen if the creator |
| // of the process maps a valid loader state when it creates the process (possibly it's trying |
| // to emulate fork, or it could be injecting libraries itself for some reason). So we also need |
| // to handle the rare case that the propery is set before startup. |
| if (dl_debug_addr_) |
| return true; // Previously set. |
| |
| uintptr_t debug_addr = 0; |
| if (handle().get_property(ZX_PROP_PROCESS_DEBUG_ADDR, &debug_addr, sizeof(debug_addr)) != ZX_OK || |
| debug_addr == 0) { |
| // Register for sets on the debug addr by setting the magic value. |
| const intptr_t kMagicValue = ZX_PROCESS_DEBUG_ADDR_BREAK_ON_SET; |
| handle().set_property(ZX_PROP_PROCESS_DEBUG_ADDR, &kMagicValue, sizeof(kMagicValue)); |
| return false; |
| } |
| if (debug_addr == ZX_PROCESS_DEBUG_ADDR_BREAK_ON_SET) |
| return false; // Still not set. |
| |
| dl_debug_addr_ = debug_addr; |
| |
| // Register a breakpoint for dynamic loads. |
| if (auto load_addr = GetLoaderBreakpointAddress(process_handle(), dl_debug_addr_)) { |
| loader_breakpoint_ = std::make_unique<Breakpoint>(debug_agent_, true); |
| if (loader_breakpoint_->SetSettings("Internal shared library load breakpoint", koid(), |
| load_addr) != ZX_OK) { |
| DEBUG_LOG(Process) << LogPreamble(this) << "Could not set shared library load breakpoint at " |
| << std::hex << load_addr; |
| // Continue even in the error case: we can continue with most things working even if the |
| // loader breakpoint fails for some reason. |
| } |
| } |
| |
| module_list_.Update(process_handle(), dl_debug_addr_); |
| |
| return true; |
| } |
| |
| DebuggedProcess::SpecialBreakpointResult DebuggedProcess::HandleSpecialBreakpoint( |
| ProcessBreakpoint* optional_bp) { |
| // The special Fuchsia loader breakpoint will be a hardcodeed breakpoint (so no input |
| // ProcessBreakpoint object) before we've seen the dl_debug_addr_. |
| if (!dl_debug_addr_ && !optional_bp) { |
| if (RegisterDebugState()) { |
| // The initial loader breakpoint will happen very early in the process startup so it |
| // will be single threaded. Since the one thread is already stopped, we can skip suspending |
| // the threads and just notify the client, keeping the calling one suspended. |
| SendModuleNotification(); |
| return SpecialBreakpointResult::kKeepSuspended; |
| } |
| } |
| |
| // Our special loader breakpoint is a breakpoint we've inserted for every shared library load. |
| if (optional_bp) { |
| const auto& breakpoints = optional_bp->breakpoints(); |
| if (std::find(breakpoints.begin(), breakpoints.end(), loader_breakpoint_.get()) != |
| breakpoints.end()) { |
| if (module_list_.Update(process_handle(), dl_debug_addr_)) { |
| // The debugged process could be multithreaded and have just dynamically loaded a new |
| // module. Suspend all threads so the client can resolve breakpoint addresses before |
| // continuing. |
| SuspendAndSendModulesIfKnown(); |
| return SpecialBreakpointResult::kKeepSuspended; |
| } |
| |
| // Modules haven't changed, resume. |
| return SpecialBreakpointResult::kContinue; |
| } |
| } |
| |
| // Not one of our special breakpoints. |
| return SpecialBreakpointResult::kNotSpecial; |
| } |
| |
| void DebuggedProcess::SuspendAndSendModulesIfKnown() { |
| if (dl_debug_addr_) { |
| // This process' modules can be known. Send them. |
| // |
| // Suspend all threads while the module list is being sent. The client will resume the threads |
| // once it's loaded symbols and processed breakpoints (this may take a while and we'd like to |
| // get any breakpoints as early as possible). |
| ClientSuspendAllThreads(); |
| SendModuleNotification(); |
| } |
| } |
| |
| void DebuggedProcess::SendModuleNotification() { |
| // Notify the client of any libraries. |
| debug_ipc::NotifyModules notify; |
| notify.process_koid = koid(); |
| notify.modules = module_list_.modules(); |
| notify.timestamp = zx::clock::get_monotonic().get(); |
| |
| // All threads are assumed to be stopped. |
| for (auto& [thread_koid, thread_ptr] : threads_) |
| notify.stopped_thread_koids.push_back(thread_koid); |
| |
| DEBUG_LOG(Process) << LogPreamble(this) << "Sending modules."; |
| |
| debug_ipc::MessageWriter writer; |
| debug_ipc::WriteNotifyModules(notify, &writer); |
| debug_agent_->stream()->Write(writer.MessageComplete()); |
| } |
| |
| SoftwareBreakpoint* DebuggedProcess::FindSoftwareBreakpoint(uint64_t address) const { |
| auto it = software_breakpoints_.find(address); |
| if (it == software_breakpoints_.end()) |
| return nullptr; |
| return it->second.get(); |
| } |
| |
| HardwareBreakpoint* DebuggedProcess::FindHardwareBreakpoint(uint64_t address) const { |
| auto it = hardware_breakpoints_.find(address); |
| if (it == hardware_breakpoints_.end()) |
| return nullptr; |
| return it->second.get(); |
| } |
| |
| Watchpoint* DebuggedProcess::FindWatchpoint(const debug_ipc::AddressRange& range) const { |
| auto it = watchpoints_.lower_bound(range); |
| if (it == watchpoints_.end()) |
| return nullptr; |
| |
| for (; it != watchpoints_.end(); it++) { |
| if (it->first.Contains(range)) |
| return it->second.get(); |
| } |
| |
| return nullptr; |
| } |
| |
| zx_status_t DebuggedProcess::RegisterBreakpoint(Breakpoint* bp, uint64_t address) { |
| LogRegisterBreakpoint(FROM_HERE, this, bp, address); |
| |
| switch (bp->settings().type) { |
| case debug_ipc::BreakpointType::kSoftware: |
| return RegisterSoftwareBreakpoint(bp, address); |
| case debug_ipc::BreakpointType::kHardware: |
| return RegisterHardwareBreakpoint(bp, address); |
| case debug_ipc::BreakpointType::kReadWrite: |
| case debug_ipc::BreakpointType::kWrite: |
| FX_NOTREACHED() << "Watchpoints are registered through RegisterWatchpoint."; |
| // TODO(donosoc): Reactivate once the transition is complete. |
| return ZX_ERR_INVALID_ARGS; |
| case debug_ipc::BreakpointType::kLast: |
| FX_NOTREACHED(); |
| return ZX_ERR_INVALID_ARGS; |
| } |
| |
| FX_NOTREACHED(); |
| } |
| |
| void DebuggedProcess::UnregisterBreakpoint(Breakpoint* bp, uint64_t address) { |
| DEBUG_LOG(Process) << LogPreamble(this) << "Unregistering breakpoint " << bp->settings().id |
| << " (" << bp->settings().name << ")."; |
| |
| switch (bp->settings().type) { |
| case debug_ipc::BreakpointType::kSoftware: |
| return UnregisterSoftwareBreakpoint(bp, address); |
| case debug_ipc::BreakpointType::kHardware: |
| return UnregisterHardwareBreakpoint(bp, address); |
| case debug_ipc::BreakpointType::kReadWrite: |
| case debug_ipc::BreakpointType::kWrite: |
| FX_NOTREACHED() << "Watchpoints are unregistered through UnregisterWatchpoint."; |
| return; |
| case debug_ipc::BreakpointType::kLast: |
| FX_NOTREACHED(); |
| return; |
| } |
| |
| FX_NOTREACHED(); |
| } |
| |
| zx_status_t DebuggedProcess::RegisterWatchpoint(Breakpoint* bp, |
| const debug_ipc::AddressRange& range) { |
| FX_DCHECK(debug_ipc::IsWatchpointType(bp->settings().type)) |
| << "Breakpoint type must be kWatchpoint, got: " |
| << debug_ipc::BreakpointTypeToString(bp->settings().type); |
| |
| // NOTE: Even though the watchpoint system can handle un-aligned ranges, there is no way for |
| // an exception to determine which byte access actually triggered the exception. This means |
| // that watchpoint installed and nominal ranges should be the same. |
| // |
| // We make that check here and fail early if the range is not correctly aligned. |
| auto aligned_range = AlignRange(range); |
| if (!aligned_range.has_value() || aligned_range.value() != range) |
| return ZX_ERR_INVALID_ARGS; |
| |
| auto it = watchpoints_.find(range); |
| if (it == watchpoints_.end()) { |
| auto watchpoint = std::make_unique<Watchpoint>(bp->settings().type, bp, this, range); |
| if (zx_status_t status = watchpoint->Init(); status != ZX_OK) |
| return status; |
| |
| watchpoints_[range] = std::move(watchpoint); |
| return ZX_OK; |
| } else { |
| return it->second->RegisterBreakpoint(bp); |
| } |
| } |
| |
| void DebuggedProcess::UnregisterWatchpoint(Breakpoint* bp, const debug_ipc::AddressRange& range) { |
| FX_DCHECK(debug_ipc::IsWatchpointType(bp->settings().type)) |
| << "Breakpoint type must be kWatchpoint, got: " |
| << debug_ipc::BreakpointTypeToString(bp->settings().type); |
| |
| auto it = watchpoints_.find(range); |
| if (it == watchpoints_.end()) |
| return; |
| |
| Watchpoint* watchpoint = it->second.get(); |
| bool still_used = watchpoint->UnregisterBreakpoint(bp); |
| if (!still_used) { |
| for (auto& [thread_koid, thread] : threads_) { |
| thread->WillDeleteProcessBreakpoint(watchpoint); |
| } |
| } |
| |
| watchpoints_.erase(it); |
| } |
| |
| void DebuggedProcess::EnqueueStepOver(ProcessBreakpoint* process_breakpoint, |
| DebuggedThread* thread) { |
| // Passing the thread will delete any previous queuing of the same thread. Otherwise the thread |
| // will be recusrsively waiting for itself and can never make progress. |
| PruneStepOverQueue(thread); |
| |
| StepOverTicket ticket = {}; |
| ticket.process_breakpoint = process_breakpoint->GetWeakPtr(); |
| ticket.thread = thread->GetWeakPtr(); |
| step_over_queue_.push_back(std::move(ticket)); |
| |
| DEBUG_LOG(Process) << LogPreamble(this) << "[PB: 0x" << std::hex << process_breakpoint->address() |
| << "] Enqueing thread " << std::dec << thread->koid() |
| << " for step over. Queue size: " << step_over_queue_.size(); |
| |
| // If the queue already had an element, we wait until that element is done. |
| if (step_over_queue_.size() > 1u) |
| return; |
| |
| // This is the first ticket in the queue. We start executing it immediatelly. |
| process_breakpoint->ExecuteStepOver(thread); |
| } |
| |
| void DebuggedProcess::OnBreakpointFinishedSteppingOver() { |
| { |
| // We always tell the current breakpoint to finish the stepping over after starting the new one. |
| // This will free the other suspended threads, letting the new stepping over thread continue. |
| // |
| // We need to do this *after* starting the following breakpoint, because otherwise we introduce |
| // a window where threads are unsuspeded between breakpoints. |
| auto prev_ticket = step_over_queue_.front(); |
| auto post_execute_breakpoint = fit::defer([prev_ticket = std::move(prev_ticket)]() { |
| if (!prev_ticket.is_valid()) |
| return; |
| |
| prev_ticket.process_breakpoint->StepOverCleanup(prev_ticket.thread.get()); |
| }); |
| |
| // Pop the previous ticket (the post-complete action is already set with the deferred action). |
| step_over_queue_.pop_front(); |
| |
| // If there are still elements in the queue, we execute the next one (the queue is pruned so we |
| // know the next one is valid). |
| PruneStepOverQueue(nullptr); |
| if (!step_over_queue_.empty()) { |
| auto& ticket = step_over_queue_.front(); |
| ticket.process_breakpoint->ExecuteStepOver(ticket.thread.get()); |
| return; |
| } |
| } |
| } |
| |
| void DebuggedProcess::OnProcessTerminated() { |
| DEBUG_LOG(Process) << LogPreamble(this) << "Terminating."; |
| debug_ipc::NotifyProcessExiting notify; |
| notify.process_koid = koid(); |
| notify.return_code = process_handle_->GetReturnCode(); |
| notify.timestamp = zx::clock::get_monotonic().get(); |
| |
| debug_ipc::MessageWriter writer; |
| debug_ipc::WriteNotifyProcessExiting(notify, &writer); |
| debug_agent_->stream()->Write(writer.MessageComplete()); |
| |
| debug_agent_->RemoveDebuggedProcess(koid()); |
| // "THIS" IS NOW DELETED. |
| } |
| |
| void DebuggedProcess::OnThreadStarting(std::unique_ptr<ExceptionHandle> exception) { |
| auto thread_handle = exception->GetThreadHandle(); |
| zx_koid_t thread_id = thread_handle->GetKoid(); |
| DEBUG_LOG(Process) << LogPreamble(this) << " Thread starting with koid " << thread_id; |
| |
| // Shouldn't have this thread yet. |
| FX_DCHECK(threads_.find(thread_id) == threads_.end()); |
| |
| auto new_thread = std::make_unique<DebuggedThread>(debug_agent_, this, std::move(thread_handle), |
| ThreadCreationOption::kSuspendedKeepSuspended, |
| std::move(exception)); |
| auto added = threads_.emplace(thread_id, std::move(new_thread)); |
| |
| // Notify the client. |
| added.first->second->SendThreadNotification(); |
| } |
| |
| void DebuggedProcess::OnThreadExiting(std::unique_ptr<ExceptionHandle> exception) { |
| auto excepting_thread_handle = exception->GetThreadHandle(); |
| zx_koid_t thread_id = excepting_thread_handle->GetKoid(); |
| DEBUG_LOG(Process) << LogPreamble(this) << " Thread exiting with koid " << thread_id; |
| |
| // Clean up our DebuggedThread object. |
| auto found_thread = threads_.find(thread_id); |
| if (found_thread == threads_.end()) { |
| FX_NOTREACHED(); |
| return; |
| } |
| |
| // The thread will currently be in a "Dying" state. For it to complete its |
| // lifecycle it must be resumed. |
| exception.reset(); |
| |
| threads_.erase(thread_id); |
| |
| // Notify the client. Can't call GetThreadRecord since the thread doesn't exist any more. |
| debug_ipc::NotifyThread notify; |
| notify.record.process_koid = koid(); |
| notify.record.thread_koid = thread_id; |
| notify.record.state = debug_ipc::ThreadRecord::State::kDead; |
| notify.timestamp = zx::clock::get_monotonic().get(); |
| |
| debug_ipc::MessageWriter writer; |
| debug_ipc::WriteNotifyThread(debug_ipc::MsgHeader::Type::kNotifyThreadExiting, notify, &writer); |
| debug_agent_->stream()->Write(writer.MessageComplete()); |
| } |
| |
| void DebuggedProcess::OnException(std::unique_ptr<ExceptionHandle> exception) { |
| auto excepting_thread_handle = exception->GetThreadHandle(); |
| zx_koid_t thread_id = excepting_thread_handle->GetKoid(); |
| |
| DebuggedThread* thread = GetThread(thread_id); |
| if (!thread) { |
| FX_LOGS(ERROR) << "Exception on thread " << thread_id << " which we don't know about."; |
| return; |
| } |
| |
| thread->OnException(std::move(exception)); |
| } |
| |
| void DebuggedProcess::OnAddressSpace(const debug_ipc::AddressSpaceRequest& request, |
| debug_ipc::AddressSpaceReply* reply) { |
| reply->map = process_handle_->GetAddressSpace(request.address); |
| } |
| |
| void DebuggedProcess::OnModules(debug_ipc::ModulesReply* reply) { |
| // Modules can only be read after the debug state is set. |
| if (dl_debug_addr_) { |
| // Since the client requested the modules explicitly, force update our cache in case something |
| // changed unexpectedly. |
| module_list_.Update(process_handle(), dl_debug_addr_); |
| reply->modules = module_list_.modules(); |
| } |
| } |
| |
| void DebuggedProcess::OnWriteMemory(const debug_ipc::WriteMemoryRequest& request, |
| debug_ipc::WriteMemoryReply* reply) { |
| size_t actual = 0; |
| reply->status = |
| process_handle_->WriteMemory(request.address, &request.data[0], request.data.size(), &actual); |
| if (reply->status == ZX_OK && actual != request.data.size()) |
| reply->status = ZX_ERR_IO; // Convert partial writes to errors. |
| } |
| |
| void DebuggedProcess::OnLoadInfoHandleTable(const debug_ipc::LoadInfoHandleTableRequest& request, |
| debug_ipc::LoadInfoHandleTableReply* reply) { |
| auto result = process_handle_->GetHandles(); |
| if (result.is_error()) { |
| reply->status = result.error_value(); |
| } else { |
| reply->status = ZX_OK; |
| reply->handles = std::move(std::move(result).value()); |
| } |
| } |
| |
| void DebuggedProcess::InjectThreadForTest(std::unique_ptr<DebuggedThread> thread) { |
| zx_koid_t koid = thread->koid(); |
| threads_[koid] = std::move(thread); |
| } |
| |
| std::vector<zx_koid_t> DebuggedProcess::ClientSuspendAllThreads(zx_koid_t except_thread) { |
| std::vector<zx_koid_t> suspended_thread_koids; |
| suspended_thread_koids.reserve(threads_.size()); |
| |
| // Issue the suspension order for all the threads. |
| for (auto& [thread_koid, thread] : threads_) { |
| // Do an asynchronous suspend. We'll wait for the suspension at the bottom. If there is more |
| // than one thread this allows waiting for each to complete in parallel instead of series. |
| // |
| // Here we explitly check for something already suspended, even if re-suspending it is a no-op, |
| // because we don't want to report its state as changed. |
| if (thread_koid != except_thread && !thread->is_client_suspended()) { |
| suspended_thread_koids.push_back(thread_koid); |
| thread->ClientSuspend(false); |
| } |
| } |
| |
| // Wait on the notification for each thread. |
| zx::time deadline = DebuggedThread::DefaultSuspendDeadline(); |
| for (zx_koid_t thread_koid : suspended_thread_koids) { |
| if (DebuggedThread* thread = GetThread(thread_koid)) |
| thread->thread_handle().WaitForSuspension(deadline); |
| } |
| |
| return suspended_thread_koids; |
| } |
| |
| void DebuggedProcess::OnStdout(bool close) { |
| FX_DCHECK(stdout_.valid()); |
| if (close) { |
| DEBUG_LOG(Process) << LogPreamble(this) << "stdout closed."; |
| stdout_.Reset(); |
| return; |
| } |
| |
| auto data = ReadSocketInput(&stdout_); |
| FX_DCHECK(!data.empty()); |
| DEBUG_LOG(Process) << LogPreamble(this) |
| << "Got stdout: " << std::string(data.data(), data.size()); |
| SendIO(debug_ipc::NotifyIO::Type::kStdout, std::move(data)); |
| } |
| |
| void DebuggedProcess::OnStderr(bool close) { |
| FX_DCHECK(stderr_.valid()); |
| if (close) { |
| DEBUG_LOG(Process) << LogPreamble(this) << "stderr closed."; |
| stderr_.Reset(); |
| return; |
| } |
| |
| auto data = ReadSocketInput(&stderr_); |
| FX_DCHECK(!data.empty()); |
| DEBUG_LOG(Process) << LogPreamble(this) |
| << "Got stderr: " << std::string(data.data(), data.size()); |
| SendIO(debug_ipc::NotifyIO::Type::kStderr, std::move(data)); |
| } |
| |
| void DebuggedProcess::SendIO(debug_ipc::NotifyIO::Type type, const std::vector<char>& data) { |
| // We send the IO message in chunks. |
| auto it = data.begin(); |
| size_t size = data.size(); |
| while (size > 0) { |
| size_t chunk_size = size; |
| if (chunk_size >= debug_ipc::NotifyIO::kMaxDataSize) |
| chunk_size = debug_ipc::NotifyIO::kMaxDataSize; |
| |
| auto end = it + chunk_size; |
| std::string msg(it, end); |
| |
| it = end; |
| size -= chunk_size; |
| |
| debug_ipc::NotifyIO notify; |
| notify.process_koid = koid(); |
| notify.type = type; |
| // We tell whether this is a piece of a bigger message. |
| notify.more_data_available = size > 0; |
| notify.data = std::move(msg); |
| notify.timestamp = zx::clock::get_monotonic().get(); |
| |
| debug_ipc::MessageWriter writer; |
| debug_ipc::WriteNotifyIO(notify, &writer); |
| debug_agent_->stream()->Write(writer.MessageComplete()); |
| } |
| } |
| |
| void DebuggedProcess::PruneStepOverQueue(DebuggedThread* optional_thread) { |
| std::deque<StepOverTicket> good_tickets; |
| for (auto& ticket : step_over_queue_) { |
| if (!ticket.is_valid()) |
| continue; |
| if (optional_thread && ticket.thread && ticket.thread.get() == optional_thread) |
| continue; // Delete everything from this thread. |
| good_tickets.push_back(std::move(ticket)); |
| } |
| |
| step_over_queue_ = std::move(good_tickets); |
| } |
| |
| zx_status_t DebuggedProcess::RegisterSoftwareBreakpoint(Breakpoint* bp, uint64_t address) { |
| auto found = software_breakpoints_.find(address); |
| if (found == software_breakpoints_.end()) { |
| auto breakpoint = std::make_unique<SoftwareBreakpoint>(bp, this, address); |
| if (zx_status_t status = breakpoint->Init(); status != ZX_OK) |
| return status; |
| |
| software_breakpoints_[address] = std::move(breakpoint); |
| return ZX_OK; |
| } else { |
| return found->second->RegisterBreakpoint(bp); |
| } |
| } |
| |
| void DebuggedProcess::UnregisterSoftwareBreakpoint(Breakpoint* bp, uint64_t address) { |
| auto found = software_breakpoints_.find(address); |
| if (found == software_breakpoints_.end()) { |
| return; |
| } |
| |
| bool still_used = found->second->UnregisterBreakpoint(bp); |
| if (!still_used) { |
| for (auto& pair : threads_) |
| pair.second->WillDeleteProcessBreakpoint(found->second.get()); |
| software_breakpoints_.erase(found); |
| } |
| } |
| |
| zx_status_t DebuggedProcess::RegisterHardwareBreakpoint(Breakpoint* bp, uint64_t address) { |
| auto found = hardware_breakpoints_.find(address); |
| if (found == hardware_breakpoints_.end()) { |
| auto breakpoint = std::make_unique<HardwareBreakpoint>(bp, this, address); |
| if (zx_status_t status = breakpoint->Init(); status != ZX_OK) |
| return status; |
| |
| hardware_breakpoints_[address] = std::move(breakpoint); |
| return ZX_OK; |
| } else { |
| return found->second->RegisterBreakpoint(bp); |
| } |
| } |
| |
| void DebuggedProcess::UnregisterHardwareBreakpoint(Breakpoint* bp, uint64_t address) { |
| auto found = hardware_breakpoints_.find(address); |
| if (found == hardware_breakpoints_.end()) { |
| return; |
| } |
| |
| bool still_used = found->second->UnregisterBreakpoint(bp); |
| if (!still_used) { |
| for (auto& pair : threads_) |
| pair.second->WillDeleteProcessBreakpoint(found->second.get()); |
| hardware_breakpoints_.erase(found); |
| } |
| } |
| |
| } // namespace debug_agent |