blob: 7a107b5098c43701f60667621f110fdab0c0851b [file] [log] [blame]
// 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 <errno.h>
#include <fcntl.h>
#include <fuchsia/io/llcpp/fidl.h>
#include <lib/async-loop/cpp/loop.h>
#include <lib/async-loop/default.h>
#include <lib/debugdata/debugdata.h>
#include <lib/fdio/directory.h>
#include <lib/fdio/io.h>
#include <lib/fdio/namespace.h>
#include <lib/fdio/spawn.h>
#include <lib/fidl-async/cpp/bind.h>
#include <lib/fzl/vmo-mapper.h>
#include <lib/zx/clock.h>
#include <lib/zx/job.h>
#include <lib/zx/process.h>
#include <libgen.h>
#include <stdio.h>
#include <string.h>
#include <sys/stat.h>
#include <unistd.h>
#include <zircon/dlfcn.h>
#include <zircon/process.h>
#include <zircon/processargs.h>
#include <zircon/status.h>
#include <zircon/syscalls.h>
#include <algorithm>
#include <forward_list>
#include <functional>
#include <string>
#include <unordered_map>
#include <utility>
#include <fbl/auto_call.h>
#include <fbl/string.h>
#include <fbl/string_printf.h>
#include <fbl/unique_fd.h>
#include <fs/synchronous_vfs.h>
#include <runtests-utils/fuchsia-run-test.h>
#include <runtests-utils/profile.h>
#include <runtests-utils/service-proxy-dir.h>
namespace fio = ::llcpp::fuchsia::io;
namespace runtests {
namespace {
// Path to helper binary which can run test as a component. This binary takes
// component url as its parameter.
constexpr char kRunTestComponentPath[] = "/bin/run-test-component";
// Path to helper binary which can run test as a v2 component. This binary takes
// component url as its parameter.
constexpr char kRunTestSuitePath[] = "/bin/run-test-suite";
fbl::String RootName(const fbl::String& path) {
const size_t i = strspn(path.c_str(), "/");
const char* start = &path.c_str()[i];
const char* end = strchr(start, '/');
if (end == nullptr) {
end = &path.c_str()[path.size()];
}
return fbl::String::Concat({"/", fbl::String(start, end - start)});
}
std::error_code ReadFile(const fbl::unique_fd& fd, uint8_t* data, size_t size) {
auto* buf = data;
ssize_t count = size;
off_t off = 0;
while (count > 0) {
ssize_t len = pread(fd.get(), buf, count, off);
if (len <= 0) {
return std::error_code{errno, std::generic_category()};
}
buf += len;
count -= len;
off += len;
}
return std::error_code{};
}
std::error_code WriteFile(const fbl::unique_fd& fd, const uint8_t* data, size_t size) {
auto* buf = data;
ssize_t count = size;
off_t off = 0;
while (count > 0) {
ssize_t len = pwrite(fd.get(), buf, count, off);
if (len <= 0) {
return std::error_code{errno, std::generic_category()};
}
buf += len;
count -= len;
off += len;
}
return std::error_code{};
}
std::optional<std::string> GetVMOName(const zx::vmo& vmo) {
char name[ZX_MAX_NAME_LEN];
zx_status_t status = vmo.get_property(ZX_PROP_NAME, name, sizeof(name));
if (status != ZX_OK || name[0] == '\0') {
zx_info_handle_basic_t info;
status = vmo.get_info(ZX_INFO_HANDLE_BASIC, &info, sizeof(info), nullptr, nullptr);
if (status != ZX_OK) {
return {};
}
snprintf(name, sizeof(name), "unnamed.%" PRIu64, info.koid);
}
return name;
}
// This function processes all raw profiles that were published via data sink
// in an efficient manner. Concretely, rather than writing each data sink into
// a separate file, it merges all profiles from the same binary into a single
// profile. First it groups all VMOs by name which uniquely identifies each
// binary. Then it merges together all VMOs for the same binary together with
// data that's already on the disk (if it exists). Finally it writes the data
// back to disk (or creates the file if necessary). This ensures that at the
// end, we have exactly one profile for each binary in total, and each profile
// is read and written at most once per call to ProcessProfiles.
std::optional<std::vector<DumpFile>> ProcessProfiles(const std::vector<zx::vmo>& data,
const fbl::unique_fd& data_sink_dir_fd) {
zx_status_t status;
if (mkdirat(data_sink_dir_fd.get(), kProfileSink, 0777) != 0 && errno != EEXIST) {
fprintf(stderr, "FAILURE: cannot mkdir \"%s\" for data-sink: %s\n", kProfileSink,
strerror(errno));
return {};
}
fbl::unique_fd sink_dir_fd{openat(data_sink_dir_fd.get(), kProfileSink, O_RDONLY | O_DIRECTORY)};
if (!sink_dir_fd) {
fprintf(stderr, "FAILURE: cannot open data-sink directory \"%s\": %s\n", kProfileSink,
strerror(errno));
return {};
}
std::unordered_map<std::string, std::forward_list<std::reference_wrapper<const zx::vmo>>>
profiles;
std::vector<DumpFile> dump_files;
// Group data by profile name. The name is a hash computed from profile metadata and
// should be unique across all binaries (modulo hash collisions).
for (const auto& vmo : data) {
auto name = GetVMOName(vmo);
if (!name) {
fprintf(stderr, "FAILURE: Cannot get a name for the VMO\n");
return {};
}
profiles[*name].push_front(std::cref(vmo));
}
for (auto& [name, vmos] : profiles) {
fbl::unique_fd fd{openat(sink_dir_fd.get(), name.c_str(), O_RDWR | O_CREAT, 0666)};
if (!fd) {
fprintf(stderr, "FAILURE: Cannot open data-sink file \"%s\": %s\n", name.c_str(),
strerror(errno));
return {};
}
uint64_t buffer_size;
std::unique_ptr<uint8_t[]> buffer;
struct stat stat;
if (fstat(fd.get(), &stat) == -1) {
fprintf(stderr, "FAILURE: Cannot stat data-sink file \"%s\": %s\n", name.c_str(),
strerror(errno));
return {};
}
if (auto file_size = static_cast<uint64_t>(stat.st_size); file_size > 0) {
// The file already exists, use it to initialize the buffer...
buffer_size = file_size;
buffer = std::make_unique<uint8_t[]>(buffer_size);
if (std::error_code ec = ReadFile(fd, buffer.get(), file_size); ec) {
fprintf(stderr, "FAILURE: Cannot read data from \"%s\": %s\n", name.c_str(),
strerror(ec.value()));
return {};
}
}
while (!vmos.empty()) {
// Merge all VMOs into the buffer.
const zx::vmo& vmo = vmos.front();
vmos.pop_front();
uint64_t vmo_size;
status = vmo.get_size(&vmo_size);
if (status != ZX_OK) {
fprintf(stderr, "FAILURE: Cannot get size of VMO \"%s\" for data-sink \"%s\": %s\n",
name.c_str(), kProfileSink, zx_status_get_string(status));
return {};
}
fzl::VmoMapper mapper;
if (vmo_size > 0) {
zx_status_t status = mapper.Map(vmo, 0, vmo_size, ZX_VM_PERM_READ);
if (status != ZX_OK) {
fprintf(stderr, "FAILURE: Cannot map VMO \"%s\" for data-sink \"%s\": %s\n", name.c_str(),
kProfileSink, zx_status_get_string(status));
return {};
}
} else {
fprintf(stderr, "WARNING: Empty VMO \"%s\" published for data-sink \"%s\"\n", kProfileSink,
name.c_str());
continue;
}
if (likely(buffer)) {
if (buffer_size != vmo_size) {
fprintf(stderr, "FAILURE: Mismatch between content sizes for \"%s\": %lu != %lu\n",
name.c_str(), buffer_size, vmo_size);
}
ZX_ASSERT(buffer_size == vmo_size);
// Ensure that profiles are structuraly compatible.
if (!ProfilesCompatible(buffer.get(), reinterpret_cast<const uint8_t*>(mapper.start()),
buffer_size)) {
fprintf(stderr, "WARNING: Unable to merge profile data: %s\n",
"source profile file is not compatible");
continue;
}
MergeProfiles(buffer.get(), reinterpret_cast<const uint8_t*>(mapper.start()), buffer_size);
} else {
// ...Otherwise use the first non-empty VMO in the list to initialize the buffer.
buffer_size = vmo_size;
buffer = std::make_unique<uint8_t[]>(buffer_size);
memcpy(buffer.get(), mapper.start(), buffer_size);
}
}
// Write the data back to the file.
if (std::error_code ec = WriteFile(fd, buffer.get(), buffer_size); ec) {
fprintf(stderr, "FAILURE: Cannot write data to \"%s\": %s\n", name.c_str(),
strerror(ec.value()));
return {};
}
dump_files.push_back(DumpFile{name, JoinPath(kProfileSink, name).c_str()});
}
return dump_files;
}
std::optional<DumpFile> ProcessDataSinkDump(const std::string& sink_name, const zx::vmo& file_data,
const fbl::unique_fd& data_sink_dir_fd) {
zx_status_t status;
if (mkdirat(data_sink_dir_fd.get(), sink_name.c_str(), 0777) != 0 && errno != EEXIST) {
fprintf(stderr, "FAILURE: cannot mkdir \"%s\" for data-sink: %s\n", sink_name.c_str(),
strerror(errno));
return {};
}
fbl::unique_fd sink_dir_fd{
openat(data_sink_dir_fd.get(), sink_name.c_str(), O_RDONLY | O_DIRECTORY)};
if (!sink_dir_fd) {
fprintf(stderr, "FAILURE: cannot open data-sink directory \"%s\": %s\n", sink_name.c_str(),
strerror(errno));
return {};
}
auto name = GetVMOName(file_data);
if (!name) {
fprintf(stderr, "FAILURE: Cannot get a name for the VMO\n");
return {};
}
uint64_t size;
status = file_data.get_size(&size);
if (status != ZX_OK) {
fprintf(stderr, "FAILURE: Cannot get size of VMO \"%s\" for data-sink \"%s\": %s\n",
name->c_str(), sink_name.c_str(), zx_status_get_string(status));
return {};
}
fzl::VmoMapper mapper;
if (size > 0) {
zx_status_t status = mapper.Map(file_data, 0, size, ZX_VM_PERM_READ);
if (status != ZX_OK) {
fprintf(stderr, "FAILURE: Cannot map VMO \"%s\" for data-sink \"%s\": %s\n", name->c_str(),
sink_name.c_str(), zx_status_get_string(status));
return {};
}
} else {
fprintf(stderr, "WARNING: Empty VMO \"%s\" published for data-sink \"%s\"\n", name->c_str(),
sink_name.c_str());
return {};
}
zx_info_handle_basic_t info;
status = file_data.get_info(ZX_INFO_HANDLE_BASIC, &info, sizeof(info), nullptr, nullptr);
if (status != ZX_OK) {
fprintf(stderr, "FAILURE: Cannot get a basic info for VMO \"%s\": %s\n", name->c_str(),
zx_status_get_string(status));
return {};
}
char filename[ZX_MAX_NAME_LEN];
snprintf(filename, sizeof(filename), "%s.%" PRIu64, sink_name.c_str(), info.koid);
fbl::unique_fd fd{openat(sink_dir_fd.get(), filename, O_WRONLY | O_CREAT | O_EXCL, 0666)};
if (!fd) {
fprintf(stderr, "FAILURE: Cannot open data-sink file \"%s\": %s\n", filename, strerror(errno));
return {};
}
if (std::error_code ec = WriteFile(fd, reinterpret_cast<uint8_t*>(mapper.start()), size); ec) {
fprintf(stderr, "FAILURE: Cannot write data to \"%s\": %s\n", filename, strerror(ec.value()));
return {};
}
return DumpFile{*name, JoinPath(sink_name, filename).c_str()};
}
} // namespace
bool SetUpForTestComponent(const char* test_path, fbl::String* out_component_executor) {
if (IsFuchsiaPkgURI(test_path)) {
const char* last_three_chars_of_url = &(test_path[strlen(test_path) - 3]);
if (0 == strncmp(last_three_chars_of_url, "cmx", 3)) { // v1 component
*out_component_executor = kRunTestComponentPath;
} else if (0 == strncmp(last_three_chars_of_url, ".cm", 3)) { // v2
*out_component_executor = kRunTestSuitePath;
} else {
fprintf(stderr, "FAILURE: component URL has unexpected format: %s\n", test_path);
return false;
}
} else if (0 == strncmp(test_path, kPkgPrefix, strlen(kPkgPrefix))) {
fprintf(stderr, "FAILURE: Test path '%s' starts with %s, which is not supported.\n", test_path,
kPkgPrefix);
return false;
}
return true;
}
std::unique_ptr<Result> RunTest(const char* argv[], const char* output_dir,
const char* output_filename, const char* test_name,
uint64_t timeout_msec, const char* realm_label) {
// The arguments passed to fdio_spawn_etc. May be overridden.
const char** args = argv;
// calculate size of argv
size_t argc = 0;
while (argv[argc] != nullptr) {
argc++;
}
const char* path = argv[0];
fbl::String component_executor;
fbl::String realm_label_arg;
if (!SetUpForTestComponent(path, &component_executor)) {
return std::make_unique<Result>(path, FAILED_TO_LAUNCH, 0, 0);
}
const char* component_launch_args[argc + 4];
if (realm_label != nullptr) {
realm_label_arg = fbl::String::Concat({"--realm-label=", realm_label});
}
if (component_executor.length() > 0) {
// Check whether the executor is present and print a more helpful error, rather than failing
// later in the fdio_spawn_etc call.
struct stat s;
if (stat(component_executor.c_str(), &s)) {
fprintf(stderr,
"FAILURE: Cannot find '%s', cannot run %s as component."
"binary.\n",
component_executor.c_str(), path);
return std::make_unique<Result>(path, FAILED_TO_LAUNCH, 0, 0);
}
component_launch_args[0] = component_executor.c_str();
int j = 1;
if (realm_label != nullptr) {
component_launch_args[j] = realm_label_arg.c_str();
j++;
}
component_launch_args[j] = path;
j++;
for (size_t i = 1; i <= argc; i++) {
component_launch_args[j] = argv[i];
j++;
}
args = component_launch_args;
}
// Truncate the name on the left so the more important stuff on the right part of the path stays
// in the name.
const char* test_name_trunc = test_name;
size_t test_name_length = strlen(test_name_trunc);
if (test_name_length > ZX_MAX_NAME_LEN - 1) {
test_name_trunc += test_name_length - (ZX_MAX_NAME_LEN - 1);
}
fbl::Vector<fdio_spawn_action_t> fdio_actions = {
fdio_spawn_action_t{.action = FDIO_SPAWN_ACTION_SET_NAME, .name = {.data = test_name_trunc}},
};
zx_status_t status;
zx::channel svc_proxy_req;
fbl::RefPtr<ServiceProxyDir> proxy_dir;
std::unique_ptr<fs::SynchronousVfs> vfs;
// This must be declared after the vfs so that its destructor gets called before the vfs
// destructor. We do this explicitly at the end of the function in the non-error case, but in
// error cases we just rely on the destructors to clean things up.
async::Loop loop{&kAsyncLoopConfigNoAttachToCurrentThread};
std::unique_ptr<debugdata::DebugData> debug_data;
// Export the root namespace.
fdio_flat_namespace_t* flat;
if ((status = fdio_ns_export_root(&flat)) != ZX_OK) {
fprintf(stderr, "FAILURE: Cannot export root namespace: %s\n", zx_status_get_string(status));
return std::make_unique<Result>(path, FAILED_UNKNOWN, 0, 0);
}
auto auto_fdio_free_flat_ns = fbl::MakeAutoCall([&flat]() { fdio_ns_free_flat_ns(flat); });
auto action_ns_entry = [](const char* prefix, zx_handle_t handle) {
return fdio_spawn_action{.action = FDIO_SPAWN_ACTION_ADD_NS_ENTRY,
.ns = {
.prefix = prefix,
.handle = handle,
}};
};
// If |output_dir| is provided, set up the loader and debugdata services that will be
// used to capture any data published.
if (output_dir != nullptr) {
fbl::unique_fd root_dir_fd{open("/", O_RDONLY | O_DIRECTORY)};
if (!root_dir_fd) {
fprintf(stderr, "FAILURE: Could not open root directory /\n");
return std::make_unique<Result>(path, FAILED_UNKNOWN, 0, 0);
}
zx::channel svc_proxy;
status = zx::channel::create(0, &svc_proxy, &svc_proxy_req);
if (status != ZX_OK) {
fprintf(stderr, "FAILURE: Cannot create channel: %s\n", zx_status_get_string(status));
return std::make_unique<Result>(path, FAILED_UNKNOWN, 0, 0);
}
zx::channel svc_handle;
for (size_t i = 0; i < flat->count; ++i) {
if (!strcmp(flat->path[i], "/svc")) {
// Save the current /svc handle...
svc_handle.reset(flat->handle[i]);
// ...and replace it with the proxy /svc.
fdio_actions.push_back(action_ns_entry("/svc", svc_proxy_req.get()));
} else {
fdio_actions.push_back(action_ns_entry(flat->path[i], flat->handle[i]));
}
}
// Setup DebugData service implementation.
debug_data = std::make_unique<debugdata::DebugData>(std::move(root_dir_fd));
// Setup proxy dir.
proxy_dir = fbl::MakeRefCounted<ServiceProxyDir>(std::move(svc_handle));
auto node = fbl::MakeRefCounted<fs::Service>(
[dispatcher = loop.dispatcher(), debug_data = debug_data.get()](zx::channel channel) {
return fidl::BindSingleInFlightOnly(dispatcher, std::move(channel), debug_data);
});
proxy_dir->AddEntry(::llcpp::fuchsia::debugdata::DebugData::Name, node);
// Setup VFS.
vfs = std::make_unique<fs::SynchronousVfs>(loop.dispatcher());
vfs->ServeDirectory(std::move(proxy_dir), std::move(svc_proxy), fs::Rights::ReadWrite());
loop.StartThread();
} else {
for (size_t i = 0; i < flat->count; ++i) {
fdio_actions.push_back(action_ns_entry(flat->path[i], flat->handle[i]));
}
}
// If |output_filename| is provided, prepare the file descriptors that will
// be used to tee the stdout/stderr of the test into the associated file.
fbl::unique_fd fds[2];
if (output_filename != nullptr) {
int temp_fds[2] = {-1, -1};
if (pipe(temp_fds)) {
fprintf(stderr, "FAILURE: Failed to create pipe: %s\n", strerror(errno));
return std::make_unique<Result>(test_name, FAILED_TO_LAUNCH, 0, 0);
}
fds[0].reset(temp_fds[0]);
fds[1].reset(temp_fds[1]);
fdio_actions.push_back(
fdio_spawn_action{.action = FDIO_SPAWN_ACTION_CLONE_FD,
.fd = {.local_fd = fds[1].get(), .target_fd = STDOUT_FILENO}});
fdio_actions.push_back(
fdio_spawn_action{.action = FDIO_SPAWN_ACTION_TRANSFER_FD,
.fd = {.local_fd = fds[1].get(), .target_fd = STDERR_FILENO}});
}
zx::job test_job;
status = zx::job::create(*zx::job::default_job(), 0, &test_job);
if (status != ZX_OK) {
fprintf(stderr, "FAILURE: zx::job::create() returned %d\n", status);
return std::make_unique<Result>(test_name, FAILED_TO_LAUNCH, 0, 0);
}
auto auto_call_kill_job = fbl::MakeAutoCall([&test_job]() { test_job.kill(); });
status = test_job.set_property(ZX_PROP_NAME, "run-test", sizeof("run-test"));
if (status != ZX_OK) {
fprintf(stderr, "FAILURE: set_property() returned %d\n", status);
return std::make_unique<Result>(test_name, FAILED_TO_LAUNCH, 0, 0);
}
// The TEST_ROOT_DIR environment variable allows tests that could be stored in
// "/system" or "/boot" to discern where they are running, and modify paths
// accordingly.
//
// TODO(fxbug.dev/3260): The hard-coded set of prefixes is not ideal. Ideally, this
// would instead set the "root" to the parent directory of the "test/"
// subdirectory where globbing was done to collect the set of tests in
// DiscoverAndRunTests(). But then it's not clear what should happen if
// using `-f` to provide a list of paths instead of directories to glob.
const fbl::String root = RootName(path);
// |root_var| must be kept alive for |env_vars| since |env_vars| may hold
// a pointer into it.
fbl::String root_var;
fbl::Vector<const char*> env_vars;
if (root == "/system" || root == "/boot") {
for (size_t i = 0; environ[i] != nullptr; ++i) {
env_vars.push_back(environ[i]);
}
root_var = fbl::String::Concat({"TEST_ROOT_DIR=", root});
env_vars.push_back(root_var.c_str());
env_vars.push_back(nullptr);
}
const char* const* env_vars_p = !env_vars.is_empty() ? env_vars.begin() : nullptr;
fds[1].release(); // To avoid double close since fdio_spawn_etc() closes it.
zx::process process;
char err_msg[FDIO_SPAWN_ERR_MSG_MAX_LENGTH];
const zx::time start_time = zx::clock::get_monotonic();
status = fdio_spawn_etc(test_job.get(), FDIO_SPAWN_CLONE_ALL & ~FDIO_SPAWN_CLONE_NAMESPACE,
args[0], args, env_vars_p, fdio_actions.size(), fdio_actions.data(),
process.reset_and_get_address(), err_msg);
if (status != ZX_OK) {
fprintf(stderr, "FAILURE: Failed to launch %s: %d (%s): %s\n", test_name, status,
zx_status_get_string(status), err_msg);
return std::make_unique<Result>(test_name, FAILED_TO_LAUNCH, 0, 0);
}
zx::time deadline = zx::time::infinite();
if (timeout_msec) {
deadline = zx::deadline_after(zx::msec(timeout_msec));
}
// Tee output.
if (output_filename != nullptr) {
FILE* output_file = fopen(output_filename, "w");
if (output_file == nullptr) {
fprintf(stderr, "FAILURE: Could not open output file at %s: %s\n", output_filename,
strerror(errno));
return std::make_unique<Result>(test_name, FAILED_DURING_IO, 0, 0);
}
if (timeout_msec) {
// If we have a timeout, we want non-blocking reads.
// This will trigger the EAGAIN code path in the read loop.
int flags = fcntl(fds[0].get(), F_GETFL, 0);
fcntl(fds[0].get(), F_SETFL, flags | O_NONBLOCK);
}
char buf[1024];
ssize_t bytes_read = 0;
while ((bytes_read = read(fds[0].get(), buf, sizeof(buf))) != 0) {
if (bytes_read > 0) {
fwrite(buf, 1, bytes_read, output_file);
fwrite(buf, 1, bytes_read, stdout);
} else if (errno == EAGAIN) {
const zx::time now = zx::clock::get_monotonic();
if (now > deadline) {
break;
}
const zx::duration sleep_for = std::min(zx::msec(100), deadline - now);
zx::nanosleep(zx::deadline_after(sleep_for));
} else {
fprintf(stderr, "Failed to read test process' output: %s\n", strerror(errno));
break;
}
}
fflush(stdout);
fflush(stderr);
fflush(output_file);
if (fclose(output_file)) {
fprintf(stderr, "FAILURE: Could not close %s: %s\n", output_filename, strerror(errno));
return std::make_unique<Result>(test_name, FAILED_DURING_IO, 0, 0);
}
}
status = process.wait_one(ZX_PROCESS_TERMINATED, deadline, nullptr);
const zx::time end_time = zx::clock::get_monotonic();
const int64_t duration_milliseconds = (end_time - start_time).to_msecs();
if (status != ZX_OK) {
if (status == ZX_ERR_TIMED_OUT) {
fprintf(stderr, "%s timed out\n", test_name);
return std::make_unique<Result>(test_name, TIMED_OUT, 0, duration_milliseconds);
}
fprintf(stderr, "FAILURE: Failed to wait for process exiting %s: %d (%s)\n", test_name, status,
zx_status_get_string(status));
return std::make_unique<Result>(test_name, FAILED_TO_WAIT, 0, duration_milliseconds);
}
// Read the return code.
zx_info_process_t proc_info;
status = process.get_info(ZX_INFO_PROCESS, &proc_info, sizeof(proc_info), nullptr, nullptr);
if (status != ZX_OK) {
fprintf(stderr, "FAILURE: Failed to get process return code %s: %d\n", test_name, status);
return std::make_unique<Result>(test_name, FAILED_TO_RETURN_CODE, 0, duration_milliseconds);
}
// The emitted signature, eg "[runtests][PASSED] /test/name", is used by the CQ/CI testrunners to
// match test names and outcomes. Changes to this format must be matched in
// https://fuchsia.googlesource.com/fuchsia/+/HEAD/tools/testing/runtests/output.go
std::unique_ptr<Result> result;
if (proc_info.return_code == 0) {
result = std::make_unique<Result>(test_name, SUCCESS, 0, duration_milliseconds);
} else {
fprintf(stderr, "%s exited with nonzero status: %" PRId64, test_name, proc_info.return_code);
result = std::make_unique<Result>(test_name, FAILED_NONZERO_RETURN_CODE, proc_info.return_code,
duration_milliseconds);
}
if (output_dir == nullptr) {
return result;
}
// Make sure that all job processes are dead before touching any data.
auto_call_kill_job.call();
// Stop the loop.
loop.Quit();
// Wait for any unfinished work to be completed.
loop.JoinThreads();
// Run one more time until there are no unprocessed messages.
loop.ResetQuit();
loop.Run(zx::time(0));
// Tear down the the VFS.
vfs.reset();
fbl::unique_fd data_sink_dir_fd{open(output_dir, O_RDONLY | O_DIRECTORY)};
if (!data_sink_dir_fd) {
printf("FAILURE: Could not open output directory %s: %s\n", "/tmp", strerror(errno));
return result;
}
for (const auto& [sink_name, data] : debug_data->data()) {
if (sink_name == kProfileSink) {
if (auto dump_files = ProcessProfiles(data, data_sink_dir_fd)) {
result->data_sinks.emplace(sink_name, std::move(*dump_files));
}
} else {
for (const auto& file_data : data) {
if (auto dump_file = ProcessDataSinkDump(sink_name, file_data, data_sink_dir_fd)) {
result->data_sinks[sink_name].push_back(*dump_file);
} else if (result->return_code == 0) {
result->launch_status = FAILED_COLLECTING_SINK_DATA;
}
}
}
}
return result;
}
} // namespace runtests