platform_system_core/init/snapuserd_transition.cpp

439 lines
14 KiB
C++
Raw Normal View History

/*
* Copyright (C) 2020 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 "snapuserd_transition.h"
#include <sys/mman.h>
#include <sys/socket.h>
#include <sys/syscall.h>
#include <sys/xattr.h>
#include <unistd.h>
#include <filesystem>
#include <string>
#include <string_view>
#include <thread>
#include <android-base/file.h>
#include <android-base/logging.h>
#include <android-base/parseint.h>
#include <android-base/stringprintf.h>
#include <android-base/strings.h>
#include <android-base/unique_fd.h>
#include <cutils/sockets.h>
init: Wait for snapuserd before starting second stage This is a race between init process and bionic libc initialization of snapuserd. init->fork() ----------------> SecondStageMain() -> PropertyInit() | | v execveat ---> __libc_init_common() -> __system_properties_init() (snapuserd) When init process calls PropertyInit(), /dev/__properties__ directory is created. When bionic libc of snapuserd daemon invokes __system_properties_init _after_ init process PropertyInit() function is invoked, libc will try to initialize the property by reading /system/etc/selinux/plat_property_contexts. Since any reads on /system has to be served by snapuserd, this specific read from libc cannot be serviced leading to deadlock. Reproduce the race by inducing a sleep of 1500ms just before execveat() so that init process calls PropertyInit() before bionic libc initialization. This leads to deadlock immediately and with additional kernel instrumentation with debug logs confirms the failure: ====================================================== init: Relaunched snapuserd with pid: 428 ext4_file_open: SNAPUSERD: path /system/etc/selinux/plat_property_contexts - Pid: 428 comm 8 ext4_file_read_iter: SNAPUSERD for path: /system/etc/selinux/plat_property_contexts pid: 428 comm 8 [ 25.418043][ T428] ext4_file_read_iter+0x3dc/0x3e0 [ 25.423000][ T428] vfs_read+0x2e0/0x354 [ 25.426986][ T428] ksys_read+0x7c/0xec [ 25.430894][ T428] __arm64_sys_read+0x20/0x30 [ 25.435419][ T428] el0_svc_common.llvm.17612735770287389485+0xd0/0x1e0 [ 25.442095][ T428] do_el0_svc+0x28/0xa0 [ 25.446100][ T428] el0_svc+0x14/0x24 [ 25.449825][ T428] el0_sync_handler+0x88/0xec [ 25.454343][ T428] el0_sync+0x1c0/0x200 ===================================================== Fix: Before starting init second stage, we will wait for snapuserd daemon to be up and running. We do a simple probe by reading system partition. This read will eventually be serviced by daemon confirming that daemon is up and running. Furthermore, we are still in the kernel domain and sepolicy has not been enforced yet. Thus, access to these device mapper block devices are ok even though we may see audit logs. Note that daemon will re-initialize the __system_property_init() as part of WaitForSocket() call. This is subtle but important; since bionic libc initialized had failed silently, it is important that this re-initialization is done. Bug: 207298357 Test: Induce the failure by explicitly delaying the call of execveat(). With fix, no issues observed. Tested incremental OTA on pixel ~15 times. Signed-off-by: Akilesh Kailash <akailash@google.com> Change-Id: I86c2de977de052bfe9dcdc002dcbd9026601d0f3
2022-01-25 08:05:31 +01:00
#include <fs_avb/fs_avb.h>
#include <libsnapshot/snapshot.h>
#include <private/android_filesystem_config.h>
#include <procinfo/process_map.h>
#include <selinux/android.h>
#include <snapuserd/snapuserd_client.h>
#include "block_dev_initializer.h"
#include "lmkd_service.h"
#include "service_utils.h"
#include "util.h"
namespace android {
namespace init {
using namespace std::string_literals;
using android::base::unique_fd;
using android::snapshot::SnapshotManager;
using android::snapshot::SnapuserdClient;
static constexpr char kSnapuserdPath[] = "/system/bin/snapuserd";
static constexpr char kSnapuserdFirstStagePidVar[] = "FIRST_STAGE_SNAPUSERD_PID";
static constexpr char kSnapuserdFirstStageFdVar[] = "FIRST_STAGE_SNAPUSERD_FD";
snapuserd: Allow connecting to the first-stage daemon. Currently there is no socket for daemon instances launched during the selinux phase of init. We don't create any sockets due to the complexity of the required sepolicy. This workaround will allow us to create the socket with very minimal sepolicy changes. init will launch a one-off instance of snapuserd in "proxy" mode, and then the following steps will occur: 1. The proxy daemon will be given two sockets, the "normal" socket that snapuserd clients would connect to, and a "proxy" socket. 2. The proxy daemon will listen on the proxy socket. 3. The first-stage daemon will wake up and connect to the proxy daemon as a client. 4. The proxy will send the normal socket via SCM_RIGHTS, then exit. 5. The first-stage daemon can now listen and accept on the normal socket. Ordering of these events is achieved through a snapuserd.proxy_ready property. Some special-casing was needed in init to make this work. The snapuserd socket owned by snapuserd_proxy is placed into a "persist" mode so it doesn't get deleted when snapuserd_proxy exits. There's also a special case method to create a Service object around a previously existing pid. Finally, first-stage init is technically on a different updateable partition than snapuserd. Thus, we add a way to query snapuserd to see if it supports socket handoff. If it does, we communicate this information through an environment variable to second-stage init. Bug: 193833730 Test: manual test Change-Id: I1950b31028980f0138bc03578cd455eb60ea4a58
2021-07-22 06:53:28 +02:00
static constexpr char kSnapuserdFirstStageInfoVar[] = "FIRST_STAGE_SNAPUSERD_INFO";
static constexpr char kSnapuserdLabel[] = "u:object_r:snapuserd_exec:s0";
static constexpr char kSnapuserdSocketLabel[] = "u:object_r:snapuserd_socket:s0";
void LaunchFirstStageSnapuserd() {
SocketDescriptor socket_desc;
socket_desc.name = android::snapshot::kSnapuserdSocket;
socket_desc.type = SOCK_STREAM;
socket_desc.perm = 0660;
socket_desc.uid = AID_SYSTEM;
socket_desc.gid = AID_SYSTEM;
// We specify a label here even though it technically is not needed. During
// first_stage_mount there is no sepolicy loaded. Once sepolicy is loaded,
// we bypass the socket entirely.
auto socket = socket_desc.Create(kSnapuserdSocketLabel);
if (!socket.ok()) {
LOG(FATAL) << "Could not create snapuserd socket: " << socket.error();
}
pid_t pid = fork();
if (pid < 0) {
PLOG(FATAL) << "Cannot launch snapuserd; fork failed";
}
if (pid == 0) {
socket->Publish();
char arg0[] = "/system/bin/snapuserd";
char arg1[] = "-user_snapshot";
char* const argv[] = {arg0, arg1, nullptr};
if (execv(arg0, argv) < 0) {
PLOG(FATAL) << "Cannot launch snapuserd; execv failed";
}
_exit(127);
}
snapuserd: Allow connecting to the first-stage daemon. Currently there is no socket for daemon instances launched during the selinux phase of init. We don't create any sockets due to the complexity of the required sepolicy. This workaround will allow us to create the socket with very minimal sepolicy changes. init will launch a one-off instance of snapuserd in "proxy" mode, and then the following steps will occur: 1. The proxy daemon will be given two sockets, the "normal" socket that snapuserd clients would connect to, and a "proxy" socket. 2. The proxy daemon will listen on the proxy socket. 3. The first-stage daemon will wake up and connect to the proxy daemon as a client. 4. The proxy will send the normal socket via SCM_RIGHTS, then exit. 5. The first-stage daemon can now listen and accept on the normal socket. Ordering of these events is achieved through a snapuserd.proxy_ready property. Some special-casing was needed in init to make this work. The snapuserd socket owned by snapuserd_proxy is placed into a "persist" mode so it doesn't get deleted when snapuserd_proxy exits. There's also a special case method to create a Service object around a previously existing pid. Finally, first-stage init is technically on a different updateable partition than snapuserd. Thus, we add a way to query snapuserd to see if it supports socket handoff. If it does, we communicate this information through an environment variable to second-stage init. Bug: 193833730 Test: manual test Change-Id: I1950b31028980f0138bc03578cd455eb60ea4a58
2021-07-22 06:53:28 +02:00
auto client = SnapuserdClient::Connect(android::snapshot::kSnapuserdSocket, 10s);
if (!client) {
LOG(FATAL) << "Could not connect to first-stage snapuserd";
}
if (client->SupportsSecondStageSocketHandoff()) {
setenv(kSnapuserdFirstStageInfoVar, "socket", 1);
}
setenv(kSnapuserdFirstStagePidVar, std::to_string(pid).c_str(), 1);
if (!client->RemoveTransitionedDaemonIndicator()) {
LOG(ERROR) << "RemoveTransitionedDaemonIndicator failed";
}
LOG(INFO) << "Relaunched snapuserd with pid: " << pid;
}
std::optional<pid_t> GetSnapuserdFirstStagePid() {
const char* pid_str = getenv(kSnapuserdFirstStagePidVar);
if (!pid_str) {
return {};
}
int pid = 0;
if (!android::base::ParseInt(pid_str, &pid)) {
LOG(FATAL) << "Could not parse pid in environment, " << kSnapuserdFirstStagePidVar << "="
<< pid_str;
}
return {pid};
}
static void RelabelLink(const std::string& link) {
selinux_android_restorecon(link.c_str(), 0);
std::string path;
if (android::base::Readlink(link, &path)) {
selinux_android_restorecon(path.c_str(), 0);
}
}
static void RelabelDeviceMapper() {
selinux_android_restorecon("/dev/device-mapper", 0);
std::error_code ec;
for (auto& iter : std::filesystem::directory_iterator("/dev/block", ec)) {
const auto& path = iter.path();
if (android::base::StartsWith(path.string(), "/dev/block/dm-")) {
selinux_android_restorecon(path.string().c_str(), 0);
}
}
}
static std::optional<int> GetRamdiskSnapuserdFd() {
const char* fd_str = getenv(kSnapuserdFirstStageFdVar);
if (!fd_str) {
return {};
}
int fd;
if (!android::base::ParseInt(fd_str, &fd)) {
LOG(FATAL) << "Could not parse fd in environment, " << kSnapuserdFirstStageFdVar << "="
<< fd_str;
}
return {fd};
}
void RestoreconRamdiskSnapuserd(int fd) {
if (fsetxattr(fd, XATTR_NAME_SELINUX, kSnapuserdLabel, strlen(kSnapuserdLabel) + 1, 0) < 0) {
PLOG(FATAL) << "fsetxattr snapuserd failed";
}
}
SnapuserdSelinuxHelper::SnapuserdSelinuxHelper(std::unique_ptr<SnapshotManager>&& sm, pid_t old_pid)
: sm_(std::move(sm)), old_pid_(old_pid) {
// Only dm-user device names change during transitions, so the other
// devices are expected to be present.
sm_->SetUeventRegenCallback([this](const std::string& device) -> bool {
if (android::base::StartsWith(device, "/dev/dm-user/")) {
return block_dev_init_.InitDmUser(android::base::Basename(device));
}
return true;
});
}
static void LockAllSystemPages() {
bool ok = true;
auto callback = [&](const android::procinfo::MapInfo& map) -> void {
if (!ok || android::base::StartsWith(map.name, "/dev/") ||
!android::base::StartsWith(map.name, "/")) {
return;
}
auto start = reinterpret_cast<const void*>(map.start);
uint64_t len = android::procinfo::MappedFileSize(map);
if (!len) {
return;
}
if (mlock(start, len) < 0) {
PLOG(ERROR) << "\"" << map.name << "\": mlock(" << start << ", " << len
<< ") failed: pgoff = " << map.pgoff;
ok = false;
}
};
if (!android::procinfo::ReadProcessMaps(getpid(), callback) || !ok) {
LOG(FATAL) << "Could not process /proc/" << getpid() << "/maps file for init";
}
}
void SnapuserdSelinuxHelper::StartTransition() {
LOG(INFO) << "Starting SELinux transition of snapuserd";
// The restorecon path reads from /system etc, so make sure any reads have
// been cached before proceeding.
auto handle = selinux_android_file_context_handle();
if (!handle) {
LOG(FATAL) << "Could not create SELinux file context handle";
}
selinux_android_set_sehandle(handle);
// We cannot access /system after the transition, so make sure init is
// pinned in memory.
LockAllSystemPages();
argv_.emplace_back("snapuserd");
argv_.emplace_back("-no_socket");
if (!sm_->PrepareSnapuserdArgsForSelinux(&argv_)) {
LOG(FATAL) << "Could not perform selinux transition";
}
}
void SnapuserdSelinuxHelper::FinishTransition() {
RelabelLink("/dev/block/by-name/super");
RelabelDeviceMapper();
selinux_android_restorecon("/dev/null", 0);
selinux_android_restorecon("/dev/urandom", 0);
selinux_android_restorecon("/dev/kmsg", 0);
selinux_android_restorecon("/dev/dm-user", SELINUX_ANDROID_RESTORECON_RECURSE);
RelaunchFirstStageSnapuserd();
if (munlockall() < 0) {
PLOG(ERROR) << "munlockall failed";
}
}
init: Wait for snapuserd before starting second stage This is a race between init process and bionic libc initialization of snapuserd. init->fork() ----------------> SecondStageMain() -> PropertyInit() | | v execveat ---> __libc_init_common() -> __system_properties_init() (snapuserd) When init process calls PropertyInit(), /dev/__properties__ directory is created. When bionic libc of snapuserd daemon invokes __system_properties_init _after_ init process PropertyInit() function is invoked, libc will try to initialize the property by reading /system/etc/selinux/plat_property_contexts. Since any reads on /system has to be served by snapuserd, this specific read from libc cannot be serviced leading to deadlock. Reproduce the race by inducing a sleep of 1500ms just before execveat() so that init process calls PropertyInit() before bionic libc initialization. This leads to deadlock immediately and with additional kernel instrumentation with debug logs confirms the failure: ====================================================== init: Relaunched snapuserd with pid: 428 ext4_file_open: SNAPUSERD: path /system/etc/selinux/plat_property_contexts - Pid: 428 comm 8 ext4_file_read_iter: SNAPUSERD for path: /system/etc/selinux/plat_property_contexts pid: 428 comm 8 [ 25.418043][ T428] ext4_file_read_iter+0x3dc/0x3e0 [ 25.423000][ T428] vfs_read+0x2e0/0x354 [ 25.426986][ T428] ksys_read+0x7c/0xec [ 25.430894][ T428] __arm64_sys_read+0x20/0x30 [ 25.435419][ T428] el0_svc_common.llvm.17612735770287389485+0xd0/0x1e0 [ 25.442095][ T428] do_el0_svc+0x28/0xa0 [ 25.446100][ T428] el0_svc+0x14/0x24 [ 25.449825][ T428] el0_sync_handler+0x88/0xec [ 25.454343][ T428] el0_sync+0x1c0/0x200 ===================================================== Fix: Before starting init second stage, we will wait for snapuserd daemon to be up and running. We do a simple probe by reading system partition. This read will eventually be serviced by daemon confirming that daemon is up and running. Furthermore, we are still in the kernel domain and sepolicy has not been enforced yet. Thus, access to these device mapper block devices are ok even though we may see audit logs. Note that daemon will re-initialize the __system_property_init() as part of WaitForSocket() call. This is subtle but important; since bionic libc initialized had failed silently, it is important that this re-initialization is done. Bug: 207298357 Test: Induce the failure by explicitly delaying the call of execveat(). With fix, no issues observed. Tested incremental OTA on pixel ~15 times. Signed-off-by: Akilesh Kailash <akailash@google.com> Change-Id: I86c2de977de052bfe9dcdc002dcbd9026601d0f3
2022-01-25 08:05:31 +01:00
/*
* Before starting init second stage, we will wait
* for snapuserd daemon to be up and running; bionic libc
* may read /system/etc/selinux/plat_property_contexts file
* before invoking main() function. This will happen if
* init initializes property during second stage. Any access
* to /system without snapuserd daemon will lead to a deadlock.
*
* Thus, we do a simple probe by reading system partition. This
* read will eventually be serviced by daemon confirming that
* daemon is up and running. Furthermore, we are still in the kernel
* domain and sepolicy has not been enforced yet. Thus, access
* to these device mapper block devices are ok even though
* we may see audit logs.
*/
bool SnapuserdSelinuxHelper::TestSnapuserdIsReady() {
// Wait for the daemon to be fully up. Daemon will write to path
// /metadata/ota/daemon-alive-indicator only when all the threads
// are ready and attached to dm-user.
//
// This check will fail for GRF devices with vendor on Android S.
// snapuserd binary from Android S won't be able to communicate
// and hence, we will fallback and issue I/O to verify
// the presence of daemon.
auto client = std::make_unique<SnapuserdClient>();
if (!client->IsTransitionedDaemonReady()) {
LOG(ERROR) << "IsTransitionedDaemonReady failed";
}
init: Wait for snapuserd before starting second stage This is a race between init process and bionic libc initialization of snapuserd. init->fork() ----------------> SecondStageMain() -> PropertyInit() | | v execveat ---> __libc_init_common() -> __system_properties_init() (snapuserd) When init process calls PropertyInit(), /dev/__properties__ directory is created. When bionic libc of snapuserd daemon invokes __system_properties_init _after_ init process PropertyInit() function is invoked, libc will try to initialize the property by reading /system/etc/selinux/plat_property_contexts. Since any reads on /system has to be served by snapuserd, this specific read from libc cannot be serviced leading to deadlock. Reproduce the race by inducing a sleep of 1500ms just before execveat() so that init process calls PropertyInit() before bionic libc initialization. This leads to deadlock immediately and with additional kernel instrumentation with debug logs confirms the failure: ====================================================== init: Relaunched snapuserd with pid: 428 ext4_file_open: SNAPUSERD: path /system/etc/selinux/plat_property_contexts - Pid: 428 comm 8 ext4_file_read_iter: SNAPUSERD for path: /system/etc/selinux/plat_property_contexts pid: 428 comm 8 [ 25.418043][ T428] ext4_file_read_iter+0x3dc/0x3e0 [ 25.423000][ T428] vfs_read+0x2e0/0x354 [ 25.426986][ T428] ksys_read+0x7c/0xec [ 25.430894][ T428] __arm64_sys_read+0x20/0x30 [ 25.435419][ T428] el0_svc_common.llvm.17612735770287389485+0xd0/0x1e0 [ 25.442095][ T428] do_el0_svc+0x28/0xa0 [ 25.446100][ T428] el0_svc+0x14/0x24 [ 25.449825][ T428] el0_sync_handler+0x88/0xec [ 25.454343][ T428] el0_sync+0x1c0/0x200 ===================================================== Fix: Before starting init second stage, we will wait for snapuserd daemon to be up and running. We do a simple probe by reading system partition. This read will eventually be serviced by daemon confirming that daemon is up and running. Furthermore, we are still in the kernel domain and sepolicy has not been enforced yet. Thus, access to these device mapper block devices are ok even though we may see audit logs. Note that daemon will re-initialize the __system_property_init() as part of WaitForSocket() call. This is subtle but important; since bionic libc initialized had failed silently, it is important that this re-initialization is done. Bug: 207298357 Test: Induce the failure by explicitly delaying the call of execveat(). With fix, no issues observed. Tested incremental OTA on pixel ~15 times. Signed-off-by: Akilesh Kailash <akailash@google.com> Change-Id: I86c2de977de052bfe9dcdc002dcbd9026601d0f3
2022-01-25 08:05:31 +01:00
std::string dev = "/dev/block/mapper/system"s + fs_mgr_get_slot_suffix();
android::base::unique_fd fd(open(dev.c_str(), O_RDONLY | O_DIRECT));
if (fd < 0) {
PLOG(ERROR) << "open " << dev << " failed";
return false;
}
void* addr;
ssize_t page_size = getpagesize();
if (posix_memalign(&addr, page_size, page_size) < 0) {
PLOG(ERROR) << "posix_memalign with page size " << page_size;
return false;
}
std::unique_ptr<void, decltype(&::free)> buffer(addr, ::free);
int iter = 0;
while (iter < 10) {
ssize_t n = TEMP_FAILURE_RETRY(pread(fd.get(), buffer.get(), page_size, 0));
if (n < 0) {
// Wait for sometime before retry
std::this_thread::sleep_for(100ms);
} else if (n == page_size) {
return true;
} else {
LOG(ERROR) << "pread returned: " << n << " from: " << dev << " expected: " << page_size;
}
iter += 1;
}
return false;
}
void SnapuserdSelinuxHelper::RelaunchFirstStageSnapuserd() {
if (!sm_->DetachFirstStageSnapuserdForSelinux()) {
LOG(FATAL) << "Could not perform selinux transition";
}
KillFirstStageSnapuserd(old_pid_);
auto fd = GetRamdiskSnapuserdFd();
if (!fd) {
LOG(FATAL) << "Environment variable " << kSnapuserdFirstStageFdVar << " was not set!";
}
unsetenv(kSnapuserdFirstStageFdVar);
RestoreconRamdiskSnapuserd(fd.value());
pid_t pid = fork();
if (pid < 0) {
PLOG(FATAL) << "Fork to relaunch snapuserd failed";
}
if (pid > 0) {
// We don't need the descriptor anymore, and it should be closed to
// avoid leaking into subprocesses.
close(fd.value());
setenv(kSnapuserdFirstStagePidVar, std::to_string(pid).c_str(), 1);
LOG(INFO) << "Relaunched snapuserd with pid: " << pid;
init: Wait for snapuserd before starting second stage This is a race between init process and bionic libc initialization of snapuserd. init->fork() ----------------> SecondStageMain() -> PropertyInit() | | v execveat ---> __libc_init_common() -> __system_properties_init() (snapuserd) When init process calls PropertyInit(), /dev/__properties__ directory is created. When bionic libc of snapuserd daemon invokes __system_properties_init _after_ init process PropertyInit() function is invoked, libc will try to initialize the property by reading /system/etc/selinux/plat_property_contexts. Since any reads on /system has to be served by snapuserd, this specific read from libc cannot be serviced leading to deadlock. Reproduce the race by inducing a sleep of 1500ms just before execveat() so that init process calls PropertyInit() before bionic libc initialization. This leads to deadlock immediately and with additional kernel instrumentation with debug logs confirms the failure: ====================================================== init: Relaunched snapuserd with pid: 428 ext4_file_open: SNAPUSERD: path /system/etc/selinux/plat_property_contexts - Pid: 428 comm 8 ext4_file_read_iter: SNAPUSERD for path: /system/etc/selinux/plat_property_contexts pid: 428 comm 8 [ 25.418043][ T428] ext4_file_read_iter+0x3dc/0x3e0 [ 25.423000][ T428] vfs_read+0x2e0/0x354 [ 25.426986][ T428] ksys_read+0x7c/0xec [ 25.430894][ T428] __arm64_sys_read+0x20/0x30 [ 25.435419][ T428] el0_svc_common.llvm.17612735770287389485+0xd0/0x1e0 [ 25.442095][ T428] do_el0_svc+0x28/0xa0 [ 25.446100][ T428] el0_svc+0x14/0x24 [ 25.449825][ T428] el0_sync_handler+0x88/0xec [ 25.454343][ T428] el0_sync+0x1c0/0x200 ===================================================== Fix: Before starting init second stage, we will wait for snapuserd daemon to be up and running. We do a simple probe by reading system partition. This read will eventually be serviced by daemon confirming that daemon is up and running. Furthermore, we are still in the kernel domain and sepolicy has not been enforced yet. Thus, access to these device mapper block devices are ok even though we may see audit logs. Note that daemon will re-initialize the __system_property_init() as part of WaitForSocket() call. This is subtle but important; since bionic libc initialized had failed silently, it is important that this re-initialization is done. Bug: 207298357 Test: Induce the failure by explicitly delaying the call of execveat(). With fix, no issues observed. Tested incremental OTA on pixel ~15 times. Signed-off-by: Akilesh Kailash <akailash@google.com> Change-Id: I86c2de977de052bfe9dcdc002dcbd9026601d0f3
2022-01-25 08:05:31 +01:00
// Since daemon is not started as a service, we have
// to explicitly set the OOM score to default which is unkillable
std::string oom_str = std::to_string(DEFAULT_OOM_SCORE_ADJUST);
std::string oom_file = android::base::StringPrintf("/proc/%d/oom_score_adj", pid);
if (!android::base::WriteStringToFile(oom_str, oom_file)) {
PLOG(ERROR) << "couldn't write oom_score_adj to snapuserd daemon with pid: " << pid;
}
init: Wait for snapuserd before starting second stage This is a race between init process and bionic libc initialization of snapuserd. init->fork() ----------------> SecondStageMain() -> PropertyInit() | | v execveat ---> __libc_init_common() -> __system_properties_init() (snapuserd) When init process calls PropertyInit(), /dev/__properties__ directory is created. When bionic libc of snapuserd daemon invokes __system_properties_init _after_ init process PropertyInit() function is invoked, libc will try to initialize the property by reading /system/etc/selinux/plat_property_contexts. Since any reads on /system has to be served by snapuserd, this specific read from libc cannot be serviced leading to deadlock. Reproduce the race by inducing a sleep of 1500ms just before execveat() so that init process calls PropertyInit() before bionic libc initialization. This leads to deadlock immediately and with additional kernel instrumentation with debug logs confirms the failure: ====================================================== init: Relaunched snapuserd with pid: 428 ext4_file_open: SNAPUSERD: path /system/etc/selinux/plat_property_contexts - Pid: 428 comm 8 ext4_file_read_iter: SNAPUSERD for path: /system/etc/selinux/plat_property_contexts pid: 428 comm 8 [ 25.418043][ T428] ext4_file_read_iter+0x3dc/0x3e0 [ 25.423000][ T428] vfs_read+0x2e0/0x354 [ 25.426986][ T428] ksys_read+0x7c/0xec [ 25.430894][ T428] __arm64_sys_read+0x20/0x30 [ 25.435419][ T428] el0_svc_common.llvm.17612735770287389485+0xd0/0x1e0 [ 25.442095][ T428] do_el0_svc+0x28/0xa0 [ 25.446100][ T428] el0_svc+0x14/0x24 [ 25.449825][ T428] el0_sync_handler+0x88/0xec [ 25.454343][ T428] el0_sync+0x1c0/0x200 ===================================================== Fix: Before starting init second stage, we will wait for snapuserd daemon to be up and running. We do a simple probe by reading system partition. This read will eventually be serviced by daemon confirming that daemon is up and running. Furthermore, we are still in the kernel domain and sepolicy has not been enforced yet. Thus, access to these device mapper block devices are ok even though we may see audit logs. Note that daemon will re-initialize the __system_property_init() as part of WaitForSocket() call. This is subtle but important; since bionic libc initialized had failed silently, it is important that this re-initialization is done. Bug: 207298357 Test: Induce the failure by explicitly delaying the call of execveat(). With fix, no issues observed. Tested incremental OTA on pixel ~15 times. Signed-off-by: Akilesh Kailash <akailash@google.com> Change-Id: I86c2de977de052bfe9dcdc002dcbd9026601d0f3
2022-01-25 08:05:31 +01:00
if (!TestSnapuserdIsReady()) {
PLOG(FATAL) << "snapuserd daemon failed to launch";
} else {
LOG(INFO) << "snapuserd daemon is up and running";
}
return;
}
// Make sure the descriptor is gone after we exec.
if (fcntl(fd.value(), F_SETFD, FD_CLOEXEC) < 0) {
PLOG(FATAL) << "fcntl FD_CLOEXEC failed for snapuserd fd";
}
std::vector<char*> argv;
for (auto& arg : argv_) {
argv.emplace_back(arg.data());
}
argv.emplace_back(nullptr);
int rv = syscall(SYS_execveat, fd.value(), "", reinterpret_cast<char* const*>(argv.data()),
nullptr, AT_EMPTY_PATH);
if (rv < 0) {
PLOG(FATAL) << "Failed to execveat() snapuserd";
}
}
std::unique_ptr<SnapuserdSelinuxHelper> SnapuserdSelinuxHelper::CreateIfNeeded() {
if (IsRecoveryMode()) {
return nullptr;
}
auto old_pid = GetSnapuserdFirstStagePid();
if (!old_pid) {
return nullptr;
}
auto sm = SnapshotManager::NewForFirstStageMount();
if (!sm) {
LOG(FATAL) << "Unable to create SnapshotManager";
}
return std::make_unique<SnapuserdSelinuxHelper>(std::move(sm), old_pid.value());
}
void KillFirstStageSnapuserd(pid_t pid) {
if (kill(pid, SIGTERM) < 0 && errno != ESRCH) {
LOG(ERROR) << "Kill snapuserd pid failed: " << pid;
} else {
LOG(INFO) << "Sent SIGTERM to snapuserd process " << pid;
}
}
void CleanupSnapuserdSocket() {
auto socket_path = ANDROID_SOCKET_DIR "/"s + android::snapshot::kSnapuserdSocket;
if (access(socket_path.c_str(), F_OK) != 0) {
return;
}
// Tell the daemon to stop accepting connections and to gracefully exit
// once all outstanding handlers have terminated.
if (auto client = SnapuserdClient::Connect(android::snapshot::kSnapuserdSocket, 3s)) {
client->DetachSnapuserd();
}
// Unlink the socket so we can create it again in second-stage.
if (unlink(socket_path.c_str()) < 0) {
PLOG(FATAL) << "unlink " << socket_path << " failed";
}
}
void SaveRamdiskPathToSnapuserd() {
int fd = open(kSnapuserdPath, O_PATH);
if (fd < 0) {
PLOG(FATAL) << "Unable to open snapuserd: " << kSnapuserdPath;
}
auto value = std::to_string(fd);
if (setenv(kSnapuserdFirstStageFdVar, value.c_str(), 1) < 0) {
PLOG(FATAL) << "setenv failed: " << kSnapuserdFirstStageFdVar << "=" << value;
}
}
bool IsFirstStageSnapuserdRunning() {
return GetSnapuserdFirstStagePid().has_value();
}
snapuserd: Allow connecting to the first-stage daemon. Currently there is no socket for daemon instances launched during the selinux phase of init. We don't create any sockets due to the complexity of the required sepolicy. This workaround will allow us to create the socket with very minimal sepolicy changes. init will launch a one-off instance of snapuserd in "proxy" mode, and then the following steps will occur: 1. The proxy daemon will be given two sockets, the "normal" socket that snapuserd clients would connect to, and a "proxy" socket. 2. The proxy daemon will listen on the proxy socket. 3. The first-stage daemon will wake up and connect to the proxy daemon as a client. 4. The proxy will send the normal socket via SCM_RIGHTS, then exit. 5. The first-stage daemon can now listen and accept on the normal socket. Ordering of these events is achieved through a snapuserd.proxy_ready property. Some special-casing was needed in init to make this work. The snapuserd socket owned by snapuserd_proxy is placed into a "persist" mode so it doesn't get deleted when snapuserd_proxy exits. There's also a special case method to create a Service object around a previously existing pid. Finally, first-stage init is technically on a different updateable partition than snapuserd. Thus, we add a way to query snapuserd to see if it supports socket handoff. If it does, we communicate this information through an environment variable to second-stage init. Bug: 193833730 Test: manual test Change-Id: I1950b31028980f0138bc03578cd455eb60ea4a58
2021-07-22 06:53:28 +02:00
std::vector<std::string> GetSnapuserdFirstStageInfo() {
const char* pid_str = getenv(kSnapuserdFirstStageInfoVar);
if (!pid_str) {
return {};
}
return android::base::Split(pid_str, ",");
}
} // namespace init
} // namespace android