2015-07-31 21:45:25 +02:00
|
|
|
/*
|
|
|
|
* Copyright (C) 2015 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 "service.h"
|
|
|
|
|
|
|
|
#include <fcntl.h>
|
2016-11-11 02:43:47 +01:00
|
|
|
#include <inttypes.h>
|
init: Add support for ambient capabilities.
Ambient capabilities are inherited in a straightforward way across
execve(2):
"
If you are nonroot but you have a capability, you can add it to pA.
If you do so, your children get that capability in pA, pP, and pE.
For example, you can set pA = CAP_NET_BIND_SERVICE, and your
children can automatically bind low-numbered ports.
"
This will allow us to get rid of the special meaning for AID_NET_ADMIN
and AID_NET_RAW, and if desired, to reduce the use of file capabilities
(which grant capabilities to any process that can execute the file). An
additional benefit of the latter is that a single .rc file can specify
all properties for a service, without having to rely on a separate file
for file capabilities.
Ambient capabilities are supported starting with kernel 4.3 and have
been backported to all Android common kernels back to 3.10.
I chose to not use Minijail here (though I'm still using libcap) for
two reasons:
1-The Minijail code is designed to work in situations where the process
is holding any set of capabilities, so it's more complex. The situation
when forking from init allows for simpler code.
2-The way Minijail is structured right now, we would not be able to
make the required SELinux calls between UID/GID dropping and other priv
dropping code. In the future, it will make sense to add some sort of
"hook" to Minijail so that it can be used in situations where we want
to do other operations between some of the privilege-dropping
operations carried out by Minijail.
Bug: 32438163
Test: Use sample service.
Change-Id: I3226cc95769d1beacbae619cb6c6e6a5425890fb
2016-10-27 16:33:03 +02:00
|
|
|
#include <linux/securebits.h>
|
2016-04-22 00:35:09 +02:00
|
|
|
#include <sched.h>
|
|
|
|
#include <sys/prctl.h>
|
2015-07-31 21:45:25 +02:00
|
|
|
#include <sys/stat.h>
|
2016-05-19 02:36:30 +02:00
|
|
|
#include <sys/time.h>
|
2015-07-31 21:45:25 +02:00
|
|
|
#include <termios.h>
|
2015-08-12 01:37:04 +02:00
|
|
|
#include <unistd.h>
|
2015-07-31 21:45:25 +02:00
|
|
|
|
2015-12-05 07:00:26 +01:00
|
|
|
#include <android-base/file.h>
|
2017-04-07 01:30:22 +02:00
|
|
|
#include <android-base/logging.h>
|
2018-05-25 03:00:39 +02:00
|
|
|
#include <android-base/properties.h>
|
2019-09-10 19:40:47 +02:00
|
|
|
#include <android-base/scopeguard.h>
|
2015-12-05 07:00:26 +01:00
|
|
|
#include <android-base/stringprintf.h>
|
2016-06-25 00:12:21 +02:00
|
|
|
#include <android-base/strings.h>
|
2019-07-09 22:33:36 +02:00
|
|
|
#include <cutils/sockets.h>
|
2016-06-01 23:03:55 +02:00
|
|
|
#include <processgroup/processgroup.h>
|
2017-04-07 01:30:22 +02:00
|
|
|
#include <selinux/selinux.h>
|
2016-06-01 23:03:55 +02:00
|
|
|
|
2019-10-23 02:18:42 +02:00
|
|
|
#include "lmkd_service.h"
|
2019-06-26 19:46:20 +02:00
|
|
|
#include "service_list.h"
|
2015-07-31 21:45:25 +02:00
|
|
|
#include "util.h"
|
|
|
|
|
2018-02-14 01:50:08 +01:00
|
|
|
#if defined(__ANDROID__)
|
2019-05-10 14:12:15 +02:00
|
|
|
#include <ApexProperties.sysprop.h>
|
2019-11-19 23:19:40 +01:00
|
|
|
#include <android/api-level.h>
|
2018-02-14 01:50:08 +01:00
|
|
|
|
Proper mount namespace configuration for bionic
This CL fixes the design problem of the previous mechanism for providing
the bootstrap bionic and the runtime bionic to the same path.
Previously, bootstrap bionic was self-bind-mounted; i.e.
/system/bin/libc.so is bind-mounted to itself. And the runtime bionic
was bind-mounted on top of the bootstrap bionic. This has not only caused
problems like `adb sync` not working(b/122737045), but also is quite
difficult to understand due to the double-and-self mounting.
This is the new design:
Most importantly, these four are all distinct:
1) bootstrap bionic (/system/lib/bootstrap/libc.so)
2) runtime bionic (/apex/com.android.runtime/lib/bionic/libc.so)
3) mount point for 1) and 2) (/bionic/lib/libc.so)
4) symlink for 3) (/system/lib/libc.so -> /bionic/lib/libc.so)
Inside the mount namespace of the pre-apexd processes, 1) is
bind-mounted to 3). Likewise, inside the mount namespace of the
post-apexd processes, 2) is bind-mounted to 3). In other words, there is
no self-mount, and no double-mount.
Another change is that mount points are under /bionic and the legacy
paths become symlinks to the mount points. This is to make sure that
there is no bind mounts under /system, which is breaking some apps.
Finally, code for creating mount namespaces, mounting bionic, etc are
refactored to mount_namespace.cpp
Bug: 120266448
Bug: 123275379
Test: m, device boots, adb sync/push/pull works,
especially with following paths:
/bionic/lib64/libc.so
/bionic/bin/linker64
/system/lib64/bootstrap/libc.so
/system/bin/bootstrap/linker64
Change-Id: Icdfbdcc1efca540ac854d4df79e07ee61fca559f
2019-01-16 15:00:59 +01:00
|
|
|
#include "mount_namespace.h"
|
2019-11-19 23:19:40 +01:00
|
|
|
#include "selinux.h"
|
2018-02-14 01:50:08 +01:00
|
|
|
#else
|
|
|
|
#include "host_init_stubs.h"
|
|
|
|
#endif
|
|
|
|
|
2017-03-24 19:43:02 +01:00
|
|
|
using android::base::boot_clock;
|
2017-06-22 21:53:17 +02:00
|
|
|
using android::base::GetProperty;
|
|
|
|
using android::base::Join;
|
2019-09-10 19:40:47 +02:00
|
|
|
using android::base::make_scope_guard;
|
2019-08-20 00:21:25 +02:00
|
|
|
using android::base::SetProperty;
|
2017-06-22 21:53:17 +02:00
|
|
|
using android::base::StartsWith;
|
2015-08-26 20:43:36 +02:00
|
|
|
using android::base::StringPrintf;
|
|
|
|
using android::base::WriteStringToFile;
|
|
|
|
|
2017-06-22 21:53:17 +02:00
|
|
|
namespace android {
|
|
|
|
namespace init {
|
|
|
|
|
2018-04-21 01:18:12 +02:00
|
|
|
static Result<std::string> ComputeContextFromExecutable(const std::string& service_path) {
|
2016-07-08 19:32:26 +02:00
|
|
|
std::string computed_context;
|
|
|
|
|
|
|
|
char* raw_con = nullptr;
|
|
|
|
char* raw_filecon = nullptr;
|
|
|
|
|
|
|
|
if (getcon(&raw_con) == -1) {
|
2017-08-23 01:13:59 +02:00
|
|
|
return Error() << "Could not get security context";
|
2016-07-08 19:32:26 +02:00
|
|
|
}
|
|
|
|
std::unique_ptr<char> mycon(raw_con);
|
|
|
|
|
|
|
|
if (getfilecon(service_path.c_str(), &raw_filecon) == -1) {
|
2017-08-23 01:13:59 +02:00
|
|
|
return Error() << "Could not get file context";
|
2016-07-08 19:32:26 +02:00
|
|
|
}
|
|
|
|
std::unique_ptr<char> filecon(raw_filecon);
|
|
|
|
|
|
|
|
char* new_con = nullptr;
|
|
|
|
int rc = security_compute_create(mycon.get(), filecon.get(),
|
|
|
|
string_to_security_class("process"), &new_con);
|
|
|
|
if (rc == 0) {
|
|
|
|
computed_context = new_con;
|
|
|
|
free(new_con);
|
|
|
|
}
|
|
|
|
if (rc == 0 && computed_context == mycon.get()) {
|
2017-08-25 21:08:57 +02:00
|
|
|
return Error() << "File " << service_path << "(labeled \"" << filecon.get()
|
|
|
|
<< "\") has incorrect label or no domain transition from " << mycon.get()
|
|
|
|
<< " to another SELinux domain defined. Have you configured your "
|
|
|
|
"service correctly? https://source.android.com/security/selinux/"
|
|
|
|
"device-policy#label_new_services_and_address_denials";
|
2016-07-08 19:32:26 +02:00
|
|
|
}
|
|
|
|
if (rc < 0) {
|
2017-08-23 01:13:59 +02:00
|
|
|
return Error() << "Could not get process context";
|
2016-07-08 19:32:26 +02:00
|
|
|
}
|
|
|
|
return computed_context;
|
|
|
|
}
|
|
|
|
|
2018-04-17 23:48:44 +02:00
|
|
|
static bool ExpandArgsAndExecv(const std::vector<std::string>& args, bool sigstop) {
|
2016-07-08 19:32:26 +02:00
|
|
|
std::vector<std::string> expanded_args;
|
2017-09-12 01:08:54 +02:00
|
|
|
std::vector<char*> c_strings;
|
|
|
|
|
2016-07-08 19:32:26 +02:00
|
|
|
expanded_args.resize(args.size());
|
2017-09-12 01:08:54 +02:00
|
|
|
c_strings.push_back(const_cast<char*>(args[0].data()));
|
2016-07-08 19:32:26 +02:00
|
|
|
for (std::size_t i = 1; i < args.size(); ++i) {
|
2019-07-31 22:59:15 +02:00
|
|
|
auto expanded_arg = ExpandProps(args[i]);
|
2020-02-05 19:49:33 +01:00
|
|
|
if (!expanded_arg.ok()) {
|
2019-07-31 22:59:15 +02:00
|
|
|
LOG(FATAL) << args[0] << ": cannot expand arguments': " << expanded_arg.error();
|
2016-07-08 19:32:26 +02:00
|
|
|
}
|
2019-07-31 22:59:15 +02:00
|
|
|
expanded_args[i] = *expanded_arg;
|
2017-09-12 01:08:54 +02:00
|
|
|
c_strings.push_back(expanded_args[i].data());
|
2016-07-08 19:32:26 +02:00
|
|
|
}
|
2017-09-12 01:08:54 +02:00
|
|
|
c_strings.push_back(nullptr);
|
|
|
|
|
2018-04-17 23:48:44 +02:00
|
|
|
if (sigstop) {
|
|
|
|
kill(getpid(), SIGSTOP);
|
|
|
|
}
|
|
|
|
|
2017-09-12 01:08:54 +02:00
|
|
|
return execv(c_strings[0], c_strings.data()) == 0;
|
2016-07-08 19:32:26 +02:00
|
|
|
}
|
|
|
|
|
2019-08-21 19:34:31 +02:00
|
|
|
static bool AreRuntimeApexesReady() {
|
Proper mount namespace configuration for bionic
This CL fixes the design problem of the previous mechanism for providing
the bootstrap bionic and the runtime bionic to the same path.
Previously, bootstrap bionic was self-bind-mounted; i.e.
/system/bin/libc.so is bind-mounted to itself. And the runtime bionic
was bind-mounted on top of the bootstrap bionic. This has not only caused
problems like `adb sync` not working(b/122737045), but also is quite
difficult to understand due to the double-and-self mounting.
This is the new design:
Most importantly, these four are all distinct:
1) bootstrap bionic (/system/lib/bootstrap/libc.so)
2) runtime bionic (/apex/com.android.runtime/lib/bionic/libc.so)
3) mount point for 1) and 2) (/bionic/lib/libc.so)
4) symlink for 3) (/system/lib/libc.so -> /bionic/lib/libc.so)
Inside the mount namespace of the pre-apexd processes, 1) is
bind-mounted to 3). Likewise, inside the mount namespace of the
post-apexd processes, 2) is bind-mounted to 3). In other words, there is
no self-mount, and no double-mount.
Another change is that mount points are under /bionic and the legacy
paths become symlinks to the mount points. This is to make sure that
there is no bind mounts under /system, which is breaking some apps.
Finally, code for creating mount namespaces, mounting bionic, etc are
refactored to mount_namespace.cpp
Bug: 120266448
Bug: 123275379
Test: m, device boots, adb sync/push/pull works,
especially with following paths:
/bionic/lib64/libc.so
/bionic/bin/linker64
/system/lib64/bootstrap/libc.so
/system/bin/bootstrap/linker64
Change-Id: Icdfbdcc1efca540ac854d4df79e07ee61fca559f
2019-01-16 15:00:59 +01:00
|
|
|
struct stat buf;
|
2019-08-21 19:34:31 +02:00
|
|
|
return stat("/apex/com.android.art/", &buf) == 0 &&
|
|
|
|
stat("/apex/com.android.runtime/", &buf) == 0;
|
Proper mount namespace configuration for bionic
This CL fixes the design problem of the previous mechanism for providing
the bootstrap bionic and the runtime bionic to the same path.
Previously, bootstrap bionic was self-bind-mounted; i.e.
/system/bin/libc.so is bind-mounted to itself. And the runtime bionic
was bind-mounted on top of the bootstrap bionic. This has not only caused
problems like `adb sync` not working(b/122737045), but also is quite
difficult to understand due to the double-and-self mounting.
This is the new design:
Most importantly, these four are all distinct:
1) bootstrap bionic (/system/lib/bootstrap/libc.so)
2) runtime bionic (/apex/com.android.runtime/lib/bionic/libc.so)
3) mount point for 1) and 2) (/bionic/lib/libc.so)
4) symlink for 3) (/system/lib/libc.so -> /bionic/lib/libc.so)
Inside the mount namespace of the pre-apexd processes, 1) is
bind-mounted to 3). Likewise, inside the mount namespace of the
post-apexd processes, 2) is bind-mounted to 3). In other words, there is
no self-mount, and no double-mount.
Another change is that mount points are under /bionic and the legacy
paths become symlinks to the mount points. This is to make sure that
there is no bind mounts under /system, which is breaking some apps.
Finally, code for creating mount namespaces, mounting bionic, etc are
refactored to mount_namespace.cpp
Bug: 120266448
Bug: 123275379
Test: m, device boots, adb sync/push/pull works,
especially with following paths:
/bionic/lib64/libc.so
/bionic/bin/linker64
/system/lib64/bootstrap/libc.so
/system/bin/bootstrap/linker64
Change-Id: Icdfbdcc1efca540ac854d4df79e07ee61fca559f
2019-01-16 15:00:59 +01:00
|
|
|
}
|
|
|
|
|
2017-07-27 01:09:09 +02:00
|
|
|
unsigned long Service::next_start_order_ = 1;
|
2017-07-28 23:48:41 +02:00
|
|
|
bool Service::is_exec_service_running_ = false;
|
2017-07-27 01:09:09 +02:00
|
|
|
|
2017-09-13 00:58:47 +02:00
|
|
|
Service::Service(const std::string& name, Subcontext* subcontext_for_restart_commands,
|
2019-12-05 13:35:19 +01:00
|
|
|
const std::vector<std::string>& args, bool from_apex)
|
|
|
|
: Service(name, 0, 0, 0, {}, 0, "", subcontext_for_restart_commands, args, from_apex) {}
|
2015-07-31 21:45:25 +02:00
|
|
|
|
2017-03-27 19:59:11 +02:00
|
|
|
Service::Service(const std::string& name, unsigned flags, uid_t uid, gid_t gid,
|
2019-07-09 00:09:36 +02:00
|
|
|
const std::vector<gid_t>& supp_gids, int namespace_flags,
|
2019-02-06 19:45:56 +01:00
|
|
|
const std::string& seclabel, Subcontext* subcontext_for_restart_commands,
|
2019-12-05 13:35:19 +01:00
|
|
|
const std::vector<std::string>& args, bool from_apex)
|
2017-03-27 19:59:11 +02:00
|
|
|
: name_(name),
|
|
|
|
classnames_({"default"}),
|
|
|
|
flags_(flags),
|
|
|
|
pid_(0),
|
|
|
|
crash_count_(0),
|
2019-05-30 00:58:32 +02:00
|
|
|
proc_attr_{.ioprio_class = IoSchedClass_NONE,
|
|
|
|
.ioprio_pri = 0,
|
|
|
|
.uid = uid,
|
|
|
|
.gid = gid,
|
|
|
|
.supp_gids = supp_gids,
|
|
|
|
.priority = 0},
|
|
|
|
namespaces_{.flags = namespace_flags},
|
2017-03-27 19:59:11 +02:00
|
|
|
seclabel_(seclabel),
|
2018-02-14 01:24:51 +01:00
|
|
|
onrestart_(false, subcontext_for_restart_commands, "<Service '" + name + "' onrestart>", 0,
|
|
|
|
"onrestart", {}),
|
2019-10-23 02:18:42 +02:00
|
|
|
oom_score_adjust_(DEFAULT_OOM_SCORE_ADJUST),
|
2017-07-27 01:09:09 +02:00
|
|
|
start_order_(0),
|
2019-12-05 13:35:19 +01:00
|
|
|
args_(args),
|
|
|
|
from_apex_(from_apex) {}
|
2015-07-31 21:45:25 +02:00
|
|
|
|
|
|
|
void Service::NotifyStateChange(const std::string& new_state) const {
|
2017-03-28 01:27:30 +02:00
|
|
|
if ((flags_ & SVC_TEMPORARY) != 0) {
|
|
|
|
// Services created by 'exec' are temporary and don't have properties tracking their state.
|
2015-07-31 21:45:25 +02:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2017-06-23 01:50:31 +02:00
|
|
|
std::string prop_name = "init.svc." + name_;
|
2019-08-20 00:21:25 +02:00
|
|
|
SetProperty(prop_name, new_state);
|
2016-11-11 02:43:47 +01:00
|
|
|
|
|
|
|
if (new_state == "running") {
|
|
|
|
uint64_t start_ns = time_started_.time_since_epoch().count();
|
2017-08-18 19:47:46 +02:00
|
|
|
std::string boottime_property = "ro.boottime." + name_;
|
|
|
|
if (GetProperty(boottime_property, "").empty()) {
|
2019-08-20 00:21:25 +02:00
|
|
|
SetProperty(boottime_property, std::to_string(start_ns));
|
2017-08-18 19:47:46 +02:00
|
|
|
}
|
2016-11-11 02:43:47 +01:00
|
|
|
}
|
2019-07-19 20:04:01 +02:00
|
|
|
|
|
|
|
// init.svc_debug_pid.* properties are only for tests, and should not be used
|
|
|
|
// on device for security checks.
|
|
|
|
std::string pid_property = "init.svc_debug_pid." + name_;
|
|
|
|
if (new_state == "running") {
|
2019-08-20 00:21:25 +02:00
|
|
|
SetProperty(pid_property, std::to_string(pid_));
|
2019-07-19 20:04:01 +02:00
|
|
|
} else if (new_state == "stopped") {
|
2019-08-20 00:21:25 +02:00
|
|
|
SetProperty(pid_property, "");
|
2019-07-19 20:04:01 +02:00
|
|
|
}
|
2015-07-31 21:45:25 +02:00
|
|
|
}
|
|
|
|
|
2019-11-19 23:19:40 +01:00
|
|
|
void Service::KillProcessGroup(int signal, bool report_oneshot) {
|
2017-05-04 20:32:36 +02:00
|
|
|
// If we've already seen a successful result from killProcessGroup*(), then we have removed
|
|
|
|
// the cgroup already and calling these functions a second time will simply result in an error.
|
|
|
|
// This is true regardless of which signal was sent.
|
|
|
|
// These functions handle their own logging, so no additional logging is needed.
|
|
|
|
if (!process_cgroup_empty_) {
|
|
|
|
LOG(INFO) << "Sending signal " << signal << " to service '" << name_ << "' (pid " << pid_
|
|
|
|
<< ") process group...";
|
2019-11-19 23:19:40 +01:00
|
|
|
int max_processes = 0;
|
2017-05-04 20:32:36 +02:00
|
|
|
int r;
|
|
|
|
if (signal == SIGTERM) {
|
2019-11-19 23:19:40 +01:00
|
|
|
r = killProcessGroupOnce(proc_attr_.uid, pid_, signal, &max_processes);
|
2017-05-04 20:32:36 +02:00
|
|
|
} else {
|
2019-11-19 23:19:40 +01:00
|
|
|
r = killProcessGroup(proc_attr_.uid, pid_, signal, &max_processes);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (report_oneshot && max_processes > 0) {
|
|
|
|
LOG(WARNING)
|
|
|
|
<< "Killed " << max_processes
|
|
|
|
<< " additional processes from a oneshot process group for service '" << name_
|
|
|
|
<< "'. This is new behavior, previously child processes would not be killed in "
|
|
|
|
"this case.";
|
2017-05-04 20:32:36 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
if (r == 0) process_cgroup_empty_ = true;
|
|
|
|
}
|
2019-10-23 02:18:42 +02:00
|
|
|
|
|
|
|
if (oom_score_adjust_ != DEFAULT_OOM_SCORE_ADJUST) {
|
|
|
|
LmkdUnregister(name_, pid_);
|
|
|
|
}
|
2016-06-15 23:49:57 +02:00
|
|
|
}
|
|
|
|
|
2019-05-30 00:58:32 +02:00
|
|
|
void Service::SetProcessAttributesAndCaps() {
|
init: Add support for ambient capabilities.
Ambient capabilities are inherited in a straightforward way across
execve(2):
"
If you are nonroot but you have a capability, you can add it to pA.
If you do so, your children get that capability in pA, pP, and pE.
For example, you can set pA = CAP_NET_BIND_SERVICE, and your
children can automatically bind low-numbered ports.
"
This will allow us to get rid of the special meaning for AID_NET_ADMIN
and AID_NET_RAW, and if desired, to reduce the use of file capabilities
(which grant capabilities to any process that can execute the file). An
additional benefit of the latter is that a single .rc file can specify
all properties for a service, without having to rely on a separate file
for file capabilities.
Ambient capabilities are supported starting with kernel 4.3 and have
been backported to all Android common kernels back to 3.10.
I chose to not use Minijail here (though I'm still using libcap) for
two reasons:
1-The Minijail code is designed to work in situations where the process
is holding any set of capabilities, so it's more complex. The situation
when forking from init allows for simpler code.
2-The way Minijail is structured right now, we would not be able to
make the required SELinux calls between UID/GID dropping and other priv
dropping code. In the future, it will make sense to add some sort of
"hook" to Minijail so that it can be used in situations where we want
to do other operations between some of the privilege-dropping
operations carried out by Minijail.
Bug: 32438163
Test: Use sample service.
Change-Id: I3226cc95769d1beacbae619cb6c6e6a5425890fb
2016-10-27 16:33:03 +02:00
|
|
|
// Keep capabilites on uid change.
|
2019-05-30 00:58:32 +02:00
|
|
|
if (capabilities_ && proc_attr_.uid) {
|
2017-06-30 23:04:20 +02:00
|
|
|
// If Android is running in a container, some securebits might already
|
|
|
|
// be locked, so don't change those.
|
2017-07-25 23:37:21 +02:00
|
|
|
unsigned long securebits = prctl(PR_GET_SECUREBITS);
|
|
|
|
if (securebits == -1UL) {
|
2017-06-30 23:04:20 +02:00
|
|
|
PLOG(FATAL) << "prctl(PR_GET_SECUREBITS) failed for " << name_;
|
|
|
|
}
|
|
|
|
securebits |= SECBIT_KEEP_CAPS | SECBIT_KEEP_CAPS_LOCKED;
|
|
|
|
if (prctl(PR_SET_SECUREBITS, securebits) != 0) {
|
|
|
|
PLOG(FATAL) << "prctl(PR_SET_SECUREBITS) failed for " << name_;
|
init: Add support for ambient capabilities.
Ambient capabilities are inherited in a straightforward way across
execve(2):
"
If you are nonroot but you have a capability, you can add it to pA.
If you do so, your children get that capability in pA, pP, and pE.
For example, you can set pA = CAP_NET_BIND_SERVICE, and your
children can automatically bind low-numbered ports.
"
This will allow us to get rid of the special meaning for AID_NET_ADMIN
and AID_NET_RAW, and if desired, to reduce the use of file capabilities
(which grant capabilities to any process that can execute the file). An
additional benefit of the latter is that a single .rc file can specify
all properties for a service, without having to rely on a separate file
for file capabilities.
Ambient capabilities are supported starting with kernel 4.3 and have
been backported to all Android common kernels back to 3.10.
I chose to not use Minijail here (though I'm still using libcap) for
two reasons:
1-The Minijail code is designed to work in situations where the process
is holding any set of capabilities, so it's more complex. The situation
when forking from init allows for simpler code.
2-The way Minijail is structured right now, we would not be able to
make the required SELinux calls between UID/GID dropping and other priv
dropping code. In the future, it will make sense to add some sort of
"hook" to Minijail so that it can be used in situations where we want
to do other operations between some of the privilege-dropping
operations carried out by Minijail.
Bug: 32438163
Test: Use sample service.
Change-Id: I3226cc95769d1beacbae619cb6c6e6a5425890fb
2016-10-27 16:33:03 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2020-02-05 19:49:33 +01:00
|
|
|
if (auto result = SetProcessAttributes(proc_attr_); !result.ok()) {
|
2019-05-30 00:58:32 +02:00
|
|
|
LOG(FATAL) << "cannot set attribute for " << name_ << ": " << result.error();
|
2016-07-08 19:32:26 +02:00
|
|
|
}
|
2019-05-30 00:58:32 +02:00
|
|
|
|
2016-07-08 19:32:26 +02:00
|
|
|
if (!seclabel_.empty()) {
|
|
|
|
if (setexeccon(seclabel_.c_str()) < 0) {
|
2016-07-26 03:18:16 +02:00
|
|
|
PLOG(FATAL) << "cannot setexeccon('" << seclabel_ << "') for " << name_;
|
2016-07-08 19:32:26 +02:00
|
|
|
}
|
|
|
|
}
|
2019-05-30 00:58:32 +02:00
|
|
|
|
2019-02-06 19:45:56 +01:00
|
|
|
if (capabilities_) {
|
|
|
|
if (!SetCapsForExec(*capabilities_)) {
|
init: Add support for ambient capabilities.
Ambient capabilities are inherited in a straightforward way across
execve(2):
"
If you are nonroot but you have a capability, you can add it to pA.
If you do so, your children get that capability in pA, pP, and pE.
For example, you can set pA = CAP_NET_BIND_SERVICE, and your
children can automatically bind low-numbered ports.
"
This will allow us to get rid of the special meaning for AID_NET_ADMIN
and AID_NET_RAW, and if desired, to reduce the use of file capabilities
(which grant capabilities to any process that can execute the file). An
additional benefit of the latter is that a single .rc file can specify
all properties for a service, without having to rely on a separate file
for file capabilities.
Ambient capabilities are supported starting with kernel 4.3 and have
been backported to all Android common kernels back to 3.10.
I chose to not use Minijail here (though I'm still using libcap) for
two reasons:
1-The Minijail code is designed to work in situations where the process
is holding any set of capabilities, so it's more complex. The situation
when forking from init allows for simpler code.
2-The way Minijail is structured right now, we would not be able to
make the required SELinux calls between UID/GID dropping and other priv
dropping code. In the future, it will make sense to add some sort of
"hook" to Minijail so that it can be used in situations where we want
to do other operations between some of the privilege-dropping
operations carried out by Minijail.
Bug: 32438163
Test: Use sample service.
Change-Id: I3226cc95769d1beacbae619cb6c6e6a5425890fb
2016-10-27 16:33:03 +02:00
|
|
|
LOG(FATAL) << "cannot set capabilities for " << name_;
|
|
|
|
}
|
2019-05-30 00:58:32 +02:00
|
|
|
} else if (proc_attr_.uid) {
|
2017-11-17 00:52:00 +01:00
|
|
|
// Inheritable caps can be non-zero when running in a container.
|
|
|
|
if (!DropInheritableCaps()) {
|
|
|
|
LOG(FATAL) << "cannot drop inheritable caps for " << name_;
|
|
|
|
}
|
init: Add support for ambient capabilities.
Ambient capabilities are inherited in a straightforward way across
execve(2):
"
If you are nonroot but you have a capability, you can add it to pA.
If you do so, your children get that capability in pA, pP, and pE.
For example, you can set pA = CAP_NET_BIND_SERVICE, and your
children can automatically bind low-numbered ports.
"
This will allow us to get rid of the special meaning for AID_NET_ADMIN
and AID_NET_RAW, and if desired, to reduce the use of file capabilities
(which grant capabilities to any process that can execute the file). An
additional benefit of the latter is that a single .rc file can specify
all properties for a service, without having to rely on a separate file
for file capabilities.
Ambient capabilities are supported starting with kernel 4.3 and have
been backported to all Android common kernels back to 3.10.
I chose to not use Minijail here (though I'm still using libcap) for
two reasons:
1-The Minijail code is designed to work in situations where the process
is holding any set of capabilities, so it's more complex. The situation
when forking from init allows for simpler code.
2-The way Minijail is structured right now, we would not be able to
make the required SELinux calls between UID/GID dropping and other priv
dropping code. In the future, it will make sense to add some sort of
"hook" to Minijail so that it can be used in situations where we want
to do other operations between some of the privilege-dropping
operations carried out by Minijail.
Bug: 32438163
Test: Use sample service.
Change-Id: I3226cc95769d1beacbae619cb6c6e6a5425890fb
2016-10-27 16:33:03 +02:00
|
|
|
}
|
2016-07-08 19:32:26 +02:00
|
|
|
}
|
|
|
|
|
2018-04-13 19:38:57 +02:00
|
|
|
void Service::Reap(const siginfo_t& siginfo) {
|
2015-07-31 21:45:25 +02:00
|
|
|
if (!(flags_ & SVC_ONESHOT) || (flags_ & SVC_RESTART)) {
|
2019-11-19 23:19:40 +01:00
|
|
|
KillProcessGroup(SIGKILL, false);
|
|
|
|
} else {
|
|
|
|
// Legacy behavior from ~2007 until Android R: this else branch did not exist and we did not
|
|
|
|
// kill the process group in this case.
|
|
|
|
if (SelinuxGetVendorAndroidVersion() >= __ANDROID_API_R__) {
|
|
|
|
// The new behavior in Android R is to kill these process groups in all cases. The
|
|
|
|
// 'true' parameter instructions KillProcessGroup() to report a warning message where it
|
|
|
|
// detects a difference in behavior has occurred.
|
|
|
|
KillProcessGroup(SIGKILL, true);
|
|
|
|
}
|
2015-07-31 21:45:25 +02:00
|
|
|
}
|
|
|
|
|
2019-07-09 22:33:36 +02:00
|
|
|
// Remove any socket resources we may have created.
|
|
|
|
for (const auto& socket : sockets_) {
|
|
|
|
auto path = ANDROID_SOCKET_DIR "/" + socket.name;
|
|
|
|
unlink(path.c_str());
|
|
|
|
}
|
2015-07-31 21:45:25 +02:00
|
|
|
|
2018-04-13 19:38:57 +02:00
|
|
|
for (const auto& f : reap_callbacks_) {
|
|
|
|
f(siginfo);
|
|
|
|
}
|
|
|
|
|
2019-09-10 19:40:47 +02:00
|
|
|
if ((siginfo.si_code != CLD_EXITED || siginfo.si_status != 0) && on_failure_reboot_target_) {
|
|
|
|
LOG(ERROR) << "Service with 'reboot_on_failure' option failed, shutting down system.";
|
2019-11-13 01:21:20 +01:00
|
|
|
trigger_shutdown(*on_failure_reboot_target_);
|
2019-09-10 19:40:47 +02:00
|
|
|
}
|
|
|
|
|
2017-07-28 23:48:41 +02:00
|
|
|
if (flags_ & SVC_EXEC) UnSetExec();
|
|
|
|
|
|
|
|
if (flags_ & SVC_TEMPORARY) return;
|
2015-07-31 21:45:25 +02:00
|
|
|
|
|
|
|
pid_ = 0;
|
|
|
|
flags_ &= (~SVC_RUNNING);
|
2017-07-27 01:09:09 +02:00
|
|
|
start_order_ = 0;
|
2015-07-31 21:45:25 +02:00
|
|
|
|
|
|
|
// Oneshot processes go into the disabled state on exit,
|
|
|
|
// except when manually restarted.
|
Support for stopping/starting post-data-mount class subsets.
On devices that use FDE and APEX at the same time, we need to bring up a
minimal framework to be able to mount the /data partition. During this
period, a tmpfs /data filesystem is created, which doesn't contain any
of the updated APEXEs. As a consequence, all those processes will be
using the APEXes from the /system partition.
This is obviously not desired, as APEXes in /system may be old and/or
contain security issues. Additionally, it would create a difference
between FBE and FDE devices at runtime.
Ideally, we restart all processes that have started after we created the
tmpfs /data. We can't (re)start based on class names alone, because some
classes (eg 'hal') contain services that are required to start apexd
itself and that shouldn't be killed (eg the graphics HAL).
To address this, keep track of which processes are started after /data
is mounted, with a new 'mark_post_data' keyword. Additionally, create
'class_reset_post_data', which resets all services in the class that
were created after the initial /data mount, and 'class_start_post_data',
which starts all services in the class that were started after /data was
mounted.
On a device with FBE, these keywords wouldn't be used; on a device with
FDE, we'd use them to bring down the right processes after the user has
entered the correct secret, and restart them.
Bug: 118485723
Test: manually verified process list
Change-Id: I16adb776dacf1dd1feeaff9e60639b99899905eb
2019-04-23 16:26:01 +02:00
|
|
|
if ((flags_ & SVC_ONESHOT) && !(flags_ & SVC_RESTART) && !(flags_ & SVC_RESET)) {
|
2015-07-31 21:45:25 +02:00
|
|
|
flags_ |= SVC_DISABLED;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Disabled and reset processes do not get restarted automatically.
|
|
|
|
if (flags_ & (SVC_DISABLED | SVC_RESET)) {
|
|
|
|
NotifyStateChange("stopped");
|
2017-03-28 01:27:30 +02:00
|
|
|
return;
|
2015-07-31 21:45:25 +02:00
|
|
|
}
|
|
|
|
|
2019-05-10 14:12:15 +02:00
|
|
|
#if defined(__ANDROID__)
|
|
|
|
static bool is_apex_updatable = android::sysprop::ApexProperties::updatable().value_or(false);
|
|
|
|
#else
|
|
|
|
static bool is_apex_updatable = false;
|
|
|
|
#endif
|
|
|
|
const bool is_process_updatable = !pre_apexd_ && is_apex_updatable;
|
|
|
|
|
2019-04-27 22:10:35 +02:00
|
|
|
// If we crash > 4 times in 4 minutes or before boot_completed,
|
|
|
|
// reboot into bootloader or set crashing property
|
2016-11-11 02:43:47 +01:00
|
|
|
boot_clock::time_point now = boot_clock::now();
|
2019-05-10 14:12:15 +02:00
|
|
|
if (((flags_ & SVC_CRITICAL) || is_process_updatable) && !(flags_ & SVC_RESTART)) {
|
2019-04-27 22:10:35 +02:00
|
|
|
bool boot_completed = android::base::GetBoolProperty("sys.boot_completed", false);
|
|
|
|
if (now < time_crashed_ + 4min || !boot_completed) {
|
2016-11-11 02:43:47 +01:00
|
|
|
if (++crash_count_ > 4) {
|
2019-01-07 11:19:02 +01:00
|
|
|
if (flags_ & SVC_CRITICAL) {
|
|
|
|
// Aborts into bootloader
|
2019-04-27 22:10:35 +02:00
|
|
|
LOG(FATAL) << "critical process '" << name_ << "' exited 4 times "
|
|
|
|
<< (boot_completed ? "in 4 minutes" : "before boot completed");
|
2019-01-07 11:19:02 +01:00
|
|
|
} else {
|
2019-04-27 22:10:35 +02:00
|
|
|
LOG(ERROR) << "updatable process '" << name_ << "' exited 4 times "
|
|
|
|
<< (boot_completed ? "in 4 minutes" : "before boot completed");
|
2019-01-07 11:19:02 +01:00
|
|
|
// Notifies update_verifier and apexd
|
2019-12-11 00:00:24 +01:00
|
|
|
SetProperty("sys.init.updatable_crashing_process_name", name_);
|
2019-12-20 17:34:48 +01:00
|
|
|
SetProperty("sys.init.updatable_crashing", "1");
|
2019-01-07 11:19:02 +01:00
|
|
|
}
|
2015-07-31 21:45:25 +02:00
|
|
|
}
|
|
|
|
} else {
|
|
|
|
time_crashed_ = now;
|
2016-11-11 02:43:47 +01:00
|
|
|
crash_count_ = 1;
|
2015-07-31 21:45:25 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
flags_ &= (~SVC_RESTART);
|
|
|
|
flags_ |= SVC_RESTARTING;
|
|
|
|
|
|
|
|
// Execute all onrestart commands for this service.
|
|
|
|
onrestart_.ExecuteAllCommands();
|
|
|
|
|
|
|
|
NotifyStateChange("restarting");
|
2017-03-28 01:27:30 +02:00
|
|
|
return;
|
2015-07-31 21:45:25 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
void Service::DumpState() const {
|
2016-06-25 00:12:21 +02:00
|
|
|
LOG(INFO) << "service " << name_;
|
2017-06-22 21:53:17 +02:00
|
|
|
LOG(INFO) << " class '" << Join(classnames_, " ") << "'";
|
|
|
|
LOG(INFO) << " exec " << Join(args_, " ");
|
2019-07-09 22:33:36 +02:00
|
|
|
for (const auto& socket : sockets_) {
|
|
|
|
LOG(INFO) << " socket " << socket.name;
|
|
|
|
}
|
|
|
|
for (const auto& file : files_) {
|
|
|
|
LOG(INFO) << " file " << file.name;
|
|
|
|
}
|
2015-07-31 21:45:25 +02:00
|
|
|
}
|
|
|
|
|
2015-08-26 20:43:36 +02:00
|
|
|
|
2019-06-10 20:08:01 +02:00
|
|
|
Result<void> Service::ExecStart() {
|
2019-09-10 19:40:47 +02:00
|
|
|
auto reboot_on_failure = make_scope_guard([this] {
|
|
|
|
if (on_failure_reboot_target_) {
|
2019-11-13 01:21:20 +01:00
|
|
|
trigger_shutdown(*on_failure_reboot_target_);
|
2019-09-10 19:40:47 +02:00
|
|
|
}
|
|
|
|
});
|
|
|
|
|
2018-11-12 04:08:41 +01:00
|
|
|
if (is_updatable() && !ServiceList::GetInstance().IsServicesUpdated()) {
|
|
|
|
// Don't delay the service for ExecStart() as the semantic is that
|
|
|
|
// the caller might depend on the side effect of the execution.
|
|
|
|
return Error() << "Cannot start an updatable service '" << name_
|
|
|
|
<< "' before configs from APEXes are all loaded";
|
|
|
|
}
|
|
|
|
|
2017-07-28 23:48:41 +02:00
|
|
|
flags_ |= SVC_ONESHOT;
|
2017-03-28 01:27:30 +02:00
|
|
|
|
2020-02-05 19:49:33 +01:00
|
|
|
if (auto result = Start(); !result.ok()) {
|
2017-08-23 01:13:59 +02:00
|
|
|
return result;
|
2017-03-28 01:27:30 +02:00
|
|
|
}
|
2017-07-28 23:48:41 +02:00
|
|
|
|
|
|
|
flags_ |= SVC_EXEC;
|
|
|
|
is_exec_service_running_ = true;
|
|
|
|
|
2019-05-30 00:58:32 +02:00
|
|
|
LOG(INFO) << "SVC_EXEC service '" << name_ << "' pid " << pid_ << " (uid " << proc_attr_.uid
|
|
|
|
<< " gid " << proc_attr_.gid << "+" << proc_attr_.supp_gids.size() << " context "
|
2018-06-20 23:54:52 +02:00
|
|
|
<< (!seclabel_.empty() ? seclabel_ : "default") << ") started; waiting...";
|
2017-07-28 23:48:41 +02:00
|
|
|
|
2019-09-10 19:40:47 +02:00
|
|
|
reboot_on_failure.Disable();
|
2019-06-10 20:08:01 +02:00
|
|
|
return {};
|
2017-03-28 01:27:30 +02:00
|
|
|
}
|
|
|
|
|
2019-06-10 20:08:01 +02:00
|
|
|
Result<void> Service::Start() {
|
2019-09-10 19:40:47 +02:00
|
|
|
auto reboot_on_failure = make_scope_guard([this] {
|
|
|
|
if (on_failure_reboot_target_) {
|
2019-11-13 01:21:20 +01:00
|
|
|
trigger_shutdown(*on_failure_reboot_target_);
|
2019-09-10 19:40:47 +02:00
|
|
|
}
|
|
|
|
});
|
|
|
|
|
2018-11-12 04:08:41 +01:00
|
|
|
if (is_updatable() && !ServiceList::GetInstance().IsServicesUpdated()) {
|
|
|
|
ServiceList::GetInstance().DelayService(*this);
|
|
|
|
return Error() << "Cannot start an updatable service '" << name_
|
|
|
|
<< "' before configs from APEXes are all loaded. "
|
|
|
|
<< "Queued for execution.";
|
|
|
|
}
|
|
|
|
|
2017-10-26 19:43:10 +02:00
|
|
|
bool disabled = (flags_ & (SVC_DISABLED | SVC_RESET));
|
2015-07-31 21:45:25 +02:00
|
|
|
// Starting a service removes it from the disabled or reset state and
|
|
|
|
// immediately takes it out of the restarting state if it was in there.
|
|
|
|
flags_ &= (~(SVC_DISABLED|SVC_RESTARTING|SVC_RESET|SVC_RESTART|SVC_DISABLED_START));
|
|
|
|
|
|
|
|
// Running processes require no additional work --- if they're in the
|
|
|
|
// process of exiting, we've ensured that they will immediately restart
|
2017-10-26 19:43:10 +02:00
|
|
|
// on exit, unless they are ONESHOT. For ONESHOT service, if it's in
|
|
|
|
// stopping status, we just set SVC_RESTART flag so it will get restarted
|
|
|
|
// in Reap().
|
2015-07-31 21:45:25 +02:00
|
|
|
if (flags_ & SVC_RUNNING) {
|
2017-10-26 19:43:10 +02:00
|
|
|
if ((flags_ & SVC_ONESHOT) && disabled) {
|
|
|
|
flags_ |= SVC_RESTART;
|
|
|
|
}
|
2017-08-23 01:13:59 +02:00
|
|
|
// It is not an error to try to start a service that is already running.
|
2019-09-10 19:40:47 +02:00
|
|
|
reboot_on_failure.Disable();
|
2019-06-10 20:08:01 +02:00
|
|
|
return {};
|
2015-07-31 21:45:25 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
bool needs_console = (flags_ & SVC_CONSOLE);
|
2016-03-21 09:08:07 +01:00
|
|
|
if (needs_console) {
|
2019-05-30 00:58:32 +02:00
|
|
|
if (proc_attr_.console.empty()) {
|
2019-06-26 23:20:53 +02:00
|
|
|
proc_attr_.console = "/dev/" + GetProperty("ro.boot.console", "console");
|
2016-03-21 09:08:07 +01:00
|
|
|
}
|
|
|
|
|
2016-12-21 00:52:15 +01:00
|
|
|
// Make sure that open call succeeds to ensure a console driver is
|
|
|
|
// properly registered for the device node
|
2019-05-30 00:58:32 +02:00
|
|
|
int console_fd = open(proc_attr_.console.c_str(), O_RDWR | O_CLOEXEC);
|
2016-12-21 00:52:15 +01:00
|
|
|
if (console_fd < 0) {
|
2016-03-21 09:08:07 +01:00
|
|
|
flags_ |= SVC_DISABLED;
|
2019-05-30 00:58:32 +02:00
|
|
|
return ErrnoError() << "Couldn't open console '" << proc_attr_.console << "'";
|
2016-03-21 09:08:07 +01:00
|
|
|
}
|
2016-12-21 00:52:15 +01:00
|
|
|
close(console_fd);
|
2015-07-31 21:45:25 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
struct stat sb;
|
|
|
|
if (stat(args_[0].c_str(), &sb) == -1) {
|
|
|
|
flags_ |= SVC_DISABLED;
|
2017-08-23 01:13:59 +02:00
|
|
|
return ErrnoError() << "Cannot find '" << args_[0] << "'";
|
2015-07-31 21:45:25 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
std::string scon;
|
|
|
|
if (!seclabel_.empty()) {
|
|
|
|
scon = seclabel_;
|
|
|
|
} else {
|
2018-04-21 01:18:12 +02:00
|
|
|
auto result = ComputeContextFromExecutable(args_[0]);
|
2020-02-05 19:49:33 +01:00
|
|
|
if (!result.ok()) {
|
2017-08-23 01:13:59 +02:00
|
|
|
return result.error();
|
2015-07-31 21:45:25 +02:00
|
|
|
}
|
2017-08-23 01:13:59 +02:00
|
|
|
scon = *result;
|
2015-07-31 21:45:25 +02:00
|
|
|
}
|
|
|
|
|
2019-08-21 19:34:31 +02:00
|
|
|
if (!AreRuntimeApexesReady() && !pre_apexd_) {
|
|
|
|
// If this service is started before the Runtime and ART APEXes get
|
|
|
|
// available, mark it as pre-apexd one. Note that this marking is
|
|
|
|
// permanent. So for example, if the service is re-launched (e.g., due
|
|
|
|
// to crash), it is still recognized as pre-apexd... for consistency.
|
Proper mount namespace configuration for bionic
This CL fixes the design problem of the previous mechanism for providing
the bootstrap bionic and the runtime bionic to the same path.
Previously, bootstrap bionic was self-bind-mounted; i.e.
/system/bin/libc.so is bind-mounted to itself. And the runtime bionic
was bind-mounted on top of the bootstrap bionic. This has not only caused
problems like `adb sync` not working(b/122737045), but also is quite
difficult to understand due to the double-and-self mounting.
This is the new design:
Most importantly, these four are all distinct:
1) bootstrap bionic (/system/lib/bootstrap/libc.so)
2) runtime bionic (/apex/com.android.runtime/lib/bionic/libc.so)
3) mount point for 1) and 2) (/bionic/lib/libc.so)
4) symlink for 3) (/system/lib/libc.so -> /bionic/lib/libc.so)
Inside the mount namespace of the pre-apexd processes, 1) is
bind-mounted to 3). Likewise, inside the mount namespace of the
post-apexd processes, 2) is bind-mounted to 3). In other words, there is
no self-mount, and no double-mount.
Another change is that mount points are under /bionic and the legacy
paths become symlinks to the mount points. This is to make sure that
there is no bind mounts under /system, which is breaking some apps.
Finally, code for creating mount namespaces, mounting bionic, etc are
refactored to mount_namespace.cpp
Bug: 120266448
Bug: 123275379
Test: m, device boots, adb sync/push/pull works,
especially with following paths:
/bionic/lib64/libc.so
/bionic/bin/linker64
/system/lib64/bootstrap/libc.so
/system/bin/bootstrap/linker64
Change-Id: Icdfbdcc1efca540ac854d4df79e07ee61fca559f
2019-01-16 15:00:59 +01:00
|
|
|
pre_apexd_ = true;
|
|
|
|
}
|
|
|
|
|
Support for stopping/starting post-data-mount class subsets.
On devices that use FDE and APEX at the same time, we need to bring up a
minimal framework to be able to mount the /data partition. During this
period, a tmpfs /data filesystem is created, which doesn't contain any
of the updated APEXEs. As a consequence, all those processes will be
using the APEXes from the /system partition.
This is obviously not desired, as APEXes in /system may be old and/or
contain security issues. Additionally, it would create a difference
between FBE and FDE devices at runtime.
Ideally, we restart all processes that have started after we created the
tmpfs /data. We can't (re)start based on class names alone, because some
classes (eg 'hal') contain services that are required to start apexd
itself and that shouldn't be killed (eg the graphics HAL).
To address this, keep track of which processes are started after /data
is mounted, with a new 'mark_post_data' keyword. Additionally, create
'class_reset_post_data', which resets all services in the class that
were created after the initial /data mount, and 'class_start_post_data',
which starts all services in the class that were started after /data was
mounted.
On a device with FBE, these keywords wouldn't be used; on a device with
FDE, we'd use them to bring down the right processes after the user has
entered the correct secret, and restart them.
Bug: 118485723
Test: manually verified process list
Change-Id: I16adb776dacf1dd1feeaff9e60639b99899905eb
2019-04-23 16:26:01 +02:00
|
|
|
post_data_ = ServiceList::GetInstance().IsPostData();
|
|
|
|
|
2016-10-04 23:05:39 +02:00
|
|
|
LOG(INFO) << "starting service '" << name_ << "'...";
|
2015-07-31 21:45:25 +02:00
|
|
|
|
2019-09-10 23:20:35 +02:00
|
|
|
std::vector<Descriptor> descriptors;
|
|
|
|
for (const auto& socket : sockets_) {
|
2020-02-05 19:49:33 +01:00
|
|
|
if (auto result = socket.Create(scon); result.ok()) {
|
2019-09-10 23:20:35 +02:00
|
|
|
descriptors.emplace_back(std::move(*result));
|
|
|
|
} else {
|
|
|
|
LOG(INFO) << "Could not create socket '" << socket.name << "': " << result.error();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
for (const auto& file : files_) {
|
2020-02-05 19:49:33 +01:00
|
|
|
if (auto result = file.Create(); result.ok()) {
|
2019-09-10 23:20:35 +02:00
|
|
|
descriptors.emplace_back(std::move(*result));
|
|
|
|
} else {
|
|
|
|
LOG(INFO) << "Could not open file '" << file.name << "': " << result.error();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2016-04-22 00:35:09 +02:00
|
|
|
pid_t pid = -1;
|
2019-05-30 00:58:32 +02:00
|
|
|
if (namespaces_.flags) {
|
|
|
|
pid = clone(nullptr, nullptr, namespaces_.flags | SIGCHLD, nullptr);
|
2016-04-22 00:35:09 +02:00
|
|
|
} else {
|
|
|
|
pid = fork();
|
|
|
|
}
|
|
|
|
|
2015-07-31 21:45:25 +02:00
|
|
|
if (pid == 0) {
|
|
|
|
umask(077);
|
|
|
|
|
2020-02-05 19:49:33 +01:00
|
|
|
if (auto result = EnterNamespaces(namespaces_, name_, pre_apexd_); !result.ok()) {
|
2019-05-30 00:58:32 +02:00
|
|
|
LOG(FATAL) << "Service '" << name_
|
|
|
|
<< "' failed to set up namespaces: " << result.error();
|
2016-04-22 00:35:09 +02:00
|
|
|
}
|
|
|
|
|
2017-08-23 00:41:03 +02:00
|
|
|
for (const auto& [key, value] : environment_vars_) {
|
|
|
|
setenv(key.c_str(), value.c_str(), 1);
|
2015-07-31 21:45:25 +02:00
|
|
|
}
|
|
|
|
|
2019-09-10 23:20:35 +02:00
|
|
|
for (const auto& descriptor : descriptors) {
|
|
|
|
descriptor.Publish();
|
2019-07-09 22:33:36 +02:00
|
|
|
}
|
2015-07-31 21:45:25 +02:00
|
|
|
|
2020-02-05 19:49:33 +01:00
|
|
|
if (auto result = WritePidToFiles(&writepid_files_); !result.ok()) {
|
2019-05-30 00:58:32 +02:00
|
|
|
LOG(ERROR) << "failed to write pid to files: " << result.error();
|
2015-07-31 21:45:25 +02:00
|
|
|
}
|
|
|
|
|
2016-07-08 19:32:26 +02:00
|
|
|
// As requested, set our gid, supplemental gids, uid, context, and
|
|
|
|
// priority. Aborts on failure.
|
2019-05-30 00:58:32 +02:00
|
|
|
SetProcessAttributesAndCaps();
|
2015-07-31 21:45:25 +02:00
|
|
|
|
2018-04-17 23:48:44 +02:00
|
|
|
if (!ExpandArgsAndExecv(args_, sigstop_)) {
|
2019-09-27 00:24:40 +02:00
|
|
|
PLOG(ERROR) << "cannot execv('" << args_[0]
|
|
|
|
<< "'). See the 'Debugging init' section of init's README.md for tips";
|
2015-07-31 21:45:25 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
_exit(127);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (pid < 0) {
|
|
|
|
pid_ = 0;
|
2017-08-23 01:13:59 +02:00
|
|
|
return ErrnoError() << "Failed to fork";
|
2015-07-31 21:45:25 +02:00
|
|
|
}
|
|
|
|
|
2019-10-23 02:18:42 +02:00
|
|
|
if (oom_score_adjust_ != DEFAULT_OOM_SCORE_ADJUST) {
|
2017-06-23 01:50:31 +02:00
|
|
|
std::string oom_str = std::to_string(oom_score_adjust_);
|
2016-07-22 21:07:06 +02:00
|
|
|
std::string oom_file = StringPrintf("/proc/%d/oom_score_adj", pid);
|
|
|
|
if (!WriteStringToFile(oom_str, oom_file)) {
|
2019-03-08 21:34:53 +01:00
|
|
|
PLOG(ERROR) << "couldn't write oom_score_adj";
|
2016-07-22 21:07:06 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2016-11-11 02:43:47 +01:00
|
|
|
time_started_ = boot_clock::now();
|
2015-07-31 21:45:25 +02:00
|
|
|
pid_ = pid;
|
|
|
|
flags_ |= SVC_RUNNING;
|
2017-07-27 01:09:09 +02:00
|
|
|
start_order_ = next_start_order_++;
|
2017-05-04 20:32:36 +02:00
|
|
|
process_cgroup_empty_ = false;
|
2016-06-15 23:49:57 +02:00
|
|
|
|
2018-10-30 23:49:33 +01:00
|
|
|
bool use_memcg = swappiness_ != -1 || soft_limit_in_bytes_ != -1 || limit_in_bytes_ != -1 ||
|
|
|
|
limit_percent_ != -1 || !limit_property_.empty();
|
2019-05-30 00:58:32 +02:00
|
|
|
errno = -createProcessGroup(proc_attr_.uid, pid_, use_memcg);
|
2016-06-15 23:49:57 +02:00
|
|
|
if (errno != 0) {
|
2019-05-30 00:58:32 +02:00
|
|
|
PLOG(ERROR) << "createProcessGroup(" << proc_attr_.uid << ", " << pid_
|
|
|
|
<< ") failed for service '" << name_ << "'";
|
2018-10-30 23:49:33 +01:00
|
|
|
} else if (use_memcg) {
|
2017-07-17 04:38:11 +02:00
|
|
|
if (swappiness_ != -1) {
|
2019-05-30 00:58:32 +02:00
|
|
|
if (!setProcessGroupSwappiness(proc_attr_.uid, pid_, swappiness_)) {
|
2017-07-17 04:38:11 +02:00
|
|
|
PLOG(ERROR) << "setProcessGroupSwappiness failed";
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (soft_limit_in_bytes_ != -1) {
|
2019-05-30 00:58:32 +02:00
|
|
|
if (!setProcessGroupSoftLimit(proc_attr_.uid, pid_, soft_limit_in_bytes_)) {
|
2017-07-17 04:38:11 +02:00
|
|
|
PLOG(ERROR) << "setProcessGroupSoftLimit failed";
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2018-10-30 23:49:33 +01:00
|
|
|
size_t computed_limit_in_bytes = limit_in_bytes_;
|
|
|
|
if (limit_percent_ != -1) {
|
|
|
|
long page_size = sysconf(_SC_PAGESIZE);
|
|
|
|
long num_pages = sysconf(_SC_PHYS_PAGES);
|
|
|
|
if (page_size > 0 && num_pages > 0) {
|
|
|
|
size_t max_mem = SIZE_MAX;
|
|
|
|
if (size_t(num_pages) < SIZE_MAX / size_t(page_size)) {
|
|
|
|
max_mem = size_t(num_pages) * size_t(page_size);
|
|
|
|
}
|
|
|
|
computed_limit_in_bytes =
|
|
|
|
std::min(computed_limit_in_bytes, max_mem / 100 * limit_percent_);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!limit_property_.empty()) {
|
|
|
|
// This ends up overwriting computed_limit_in_bytes but only if the
|
|
|
|
// property is defined.
|
|
|
|
computed_limit_in_bytes = android::base::GetUintProperty(
|
|
|
|
limit_property_, computed_limit_in_bytes, SIZE_MAX);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (computed_limit_in_bytes != size_t(-1)) {
|
2019-05-30 00:58:32 +02:00
|
|
|
if (!setProcessGroupLimit(proc_attr_.uid, pid_, computed_limit_in_bytes)) {
|
2017-07-17 04:38:11 +02:00
|
|
|
PLOG(ERROR) << "setProcessGroupLimit failed";
|
|
|
|
}
|
|
|
|
}
|
2016-06-15 23:49:57 +02:00
|
|
|
}
|
2015-07-31 21:45:25 +02:00
|
|
|
|
2019-10-23 02:18:42 +02:00
|
|
|
if (oom_score_adjust_ != DEFAULT_OOM_SCORE_ADJUST) {
|
|
|
|
LmkdRegister(name_, proc_attr_.uid, pid_, oom_score_adjust_);
|
|
|
|
}
|
|
|
|
|
2015-07-31 21:45:25 +02:00
|
|
|
NotifyStateChange("running");
|
2019-09-10 19:40:47 +02:00
|
|
|
reboot_on_failure.Disable();
|
2019-06-10 20:08:01 +02:00
|
|
|
return {};
|
2015-07-31 21:45:25 +02:00
|
|
|
}
|
|
|
|
|
2019-06-10 20:08:01 +02:00
|
|
|
Result<void> Service::StartIfNotDisabled() {
|
2015-07-31 21:45:25 +02:00
|
|
|
if (!(flags_ & SVC_DISABLED)) {
|
|
|
|
return Start();
|
|
|
|
} else {
|
|
|
|
flags_ |= SVC_DISABLED_START;
|
|
|
|
}
|
2019-06-10 20:08:01 +02:00
|
|
|
return {};
|
2015-07-31 21:45:25 +02:00
|
|
|
}
|
|
|
|
|
2019-06-10 20:08:01 +02:00
|
|
|
Result<void> Service::Enable() {
|
2015-07-31 21:45:25 +02:00
|
|
|
flags_ &= ~(SVC_DISABLED | SVC_RC_DISABLED);
|
|
|
|
if (flags_ & SVC_DISABLED_START) {
|
|
|
|
return Start();
|
|
|
|
}
|
2019-06-10 20:08:01 +02:00
|
|
|
return {};
|
2015-07-31 21:45:25 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
void Service::Reset() {
|
|
|
|
StopOrReset(SVC_RESET);
|
|
|
|
}
|
|
|
|
|
Support for stopping/starting post-data-mount class subsets.
On devices that use FDE and APEX at the same time, we need to bring up a
minimal framework to be able to mount the /data partition. During this
period, a tmpfs /data filesystem is created, which doesn't contain any
of the updated APEXEs. As a consequence, all those processes will be
using the APEXes from the /system partition.
This is obviously not desired, as APEXes in /system may be old and/or
contain security issues. Additionally, it would create a difference
between FBE and FDE devices at runtime.
Ideally, we restart all processes that have started after we created the
tmpfs /data. We can't (re)start based on class names alone, because some
classes (eg 'hal') contain services that are required to start apexd
itself and that shouldn't be killed (eg the graphics HAL).
To address this, keep track of which processes are started after /data
is mounted, with a new 'mark_post_data' keyword. Additionally, create
'class_reset_post_data', which resets all services in the class that
were created after the initial /data mount, and 'class_start_post_data',
which starts all services in the class that were started after /data was
mounted.
On a device with FBE, these keywords wouldn't be used; on a device with
FDE, we'd use them to bring down the right processes after the user has
entered the correct secret, and restart them.
Bug: 118485723
Test: manually verified process list
Change-Id: I16adb776dacf1dd1feeaff9e60639b99899905eb
2019-04-23 16:26:01 +02:00
|
|
|
void Service::ResetIfPostData() {
|
|
|
|
if (post_data_) {
|
2019-05-15 22:04:13 +02:00
|
|
|
if (flags_ & SVC_RUNNING) {
|
|
|
|
running_at_post_data_reset_ = true;
|
|
|
|
}
|
Support for stopping/starting post-data-mount class subsets.
On devices that use FDE and APEX at the same time, we need to bring up a
minimal framework to be able to mount the /data partition. During this
period, a tmpfs /data filesystem is created, which doesn't contain any
of the updated APEXEs. As a consequence, all those processes will be
using the APEXes from the /system partition.
This is obviously not desired, as APEXes in /system may be old and/or
contain security issues. Additionally, it would create a difference
between FBE and FDE devices at runtime.
Ideally, we restart all processes that have started after we created the
tmpfs /data. We can't (re)start based on class names alone, because some
classes (eg 'hal') contain services that are required to start apexd
itself and that shouldn't be killed (eg the graphics HAL).
To address this, keep track of which processes are started after /data
is mounted, with a new 'mark_post_data' keyword. Additionally, create
'class_reset_post_data', which resets all services in the class that
were created after the initial /data mount, and 'class_start_post_data',
which starts all services in the class that were started after /data was
mounted.
On a device with FBE, these keywords wouldn't be used; on a device with
FDE, we'd use them to bring down the right processes after the user has
entered the correct secret, and restart them.
Bug: 118485723
Test: manually verified process list
Change-Id: I16adb776dacf1dd1feeaff9e60639b99899905eb
2019-04-23 16:26:01 +02:00
|
|
|
StopOrReset(SVC_RESET);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2019-06-10 20:08:01 +02:00
|
|
|
Result<void> Service::StartIfPostData() {
|
2019-05-15 22:04:13 +02:00
|
|
|
// Start the service, but only if it was started after /data was mounted,
|
|
|
|
// and it was still running when we reset the post-data services.
|
|
|
|
if (running_at_post_data_reset_) {
|
|
|
|
return Start();
|
|
|
|
}
|
|
|
|
|
2019-06-10 20:08:01 +02:00
|
|
|
return {};
|
2019-05-15 22:04:13 +02:00
|
|
|
}
|
|
|
|
|
2015-07-31 21:45:25 +02:00
|
|
|
void Service::Stop() {
|
|
|
|
StopOrReset(SVC_DISABLED);
|
|
|
|
}
|
|
|
|
|
2015-12-18 20:39:59 +01:00
|
|
|
void Service::Terminate() {
|
|
|
|
flags_ &= ~(SVC_RESTARTING | SVC_DISABLED_START);
|
|
|
|
flags_ |= SVC_DISABLED;
|
|
|
|
if (pid_) {
|
2016-06-15 23:49:57 +02:00
|
|
|
KillProcessGroup(SIGTERM);
|
2015-12-18 20:39:59 +01:00
|
|
|
NotifyStateChange("stopping");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2018-09-28 01:10:46 +02:00
|
|
|
void Service::Timeout() {
|
|
|
|
// All process state flags will be taken care of in Reap(), we really just want to kill the
|
|
|
|
// process here when it times out. Oneshot processes will transition to be disabled, and
|
|
|
|
// all other processes will transition to be restarting.
|
|
|
|
LOG(INFO) << "Service '" << name_ << "' expired its timeout of " << timeout_period_->count()
|
|
|
|
<< " seconds and will now be killed";
|
|
|
|
if (pid_) {
|
|
|
|
KillProcessGroup(SIGKILL);
|
|
|
|
NotifyStateChange("stopping");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2015-07-31 21:45:25 +02:00
|
|
|
void Service::Restart() {
|
|
|
|
if (flags_ & SVC_RUNNING) {
|
|
|
|
/* Stop, wait, then start the service. */
|
|
|
|
StopOrReset(SVC_RESTART);
|
|
|
|
} else if (!(flags_ & SVC_RESTARTING)) {
|
|
|
|
/* Just start the service since it's not running. */
|
2020-02-05 19:49:33 +01:00
|
|
|
if (auto result = Start(); !result.ok()) {
|
2017-08-23 01:13:59 +02:00
|
|
|
LOG(ERROR) << "Could not restart '" << name_ << "': " << result.error();
|
|
|
|
}
|
2015-07-31 21:45:25 +02:00
|
|
|
} /* else: Service is restarting anyways. */
|
|
|
|
}
|
|
|
|
|
2016-06-15 23:49:57 +02:00
|
|
|
// The how field should be either SVC_DISABLED, SVC_RESET, or SVC_RESTART.
|
2015-07-31 21:45:25 +02:00
|
|
|
void Service::StopOrReset(int how) {
|
2016-06-15 23:49:57 +02:00
|
|
|
// The service is still SVC_RUNNING until its process exits, but if it has
|
|
|
|
// already exited it shoudn't attempt a restart yet.
|
2015-07-31 21:45:25 +02:00
|
|
|
flags_ &= ~(SVC_RESTARTING | SVC_DISABLED_START);
|
|
|
|
|
|
|
|
if ((how != SVC_DISABLED) && (how != SVC_RESET) && (how != SVC_RESTART)) {
|
2016-06-15 23:49:57 +02:00
|
|
|
// An illegal flag: default to SVC_DISABLED.
|
2015-07-31 21:45:25 +02:00
|
|
|
how = SVC_DISABLED;
|
|
|
|
}
|
2016-06-15 23:49:57 +02:00
|
|
|
|
|
|
|
// If the service has not yet started, prevent it from auto-starting with its class.
|
2015-07-31 21:45:25 +02:00
|
|
|
if (how == SVC_RESET) {
|
|
|
|
flags_ |= (flags_ & SVC_RC_DISABLED) ? SVC_DISABLED : SVC_RESET;
|
|
|
|
} else {
|
|
|
|
flags_ |= how;
|
|
|
|
}
|
2017-10-27 20:29:13 +02:00
|
|
|
// Make sure it's in right status when a restart immediately follow a
|
|
|
|
// stop/reset or vice versa.
|
|
|
|
if (how == SVC_RESTART) {
|
|
|
|
flags_ &= (~(SVC_DISABLED | SVC_RESET));
|
|
|
|
} else {
|
|
|
|
flags_ &= (~SVC_RESTART);
|
|
|
|
}
|
2015-07-31 21:45:25 +02:00
|
|
|
|
|
|
|
if (pid_) {
|
2016-06-15 23:49:57 +02:00
|
|
|
KillProcessGroup(SIGKILL);
|
2015-07-31 21:45:25 +02:00
|
|
|
NotifyStateChange("stopping");
|
|
|
|
} else {
|
|
|
|
NotifyStateChange("stopped");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2019-07-30 18:34:41 +02:00
|
|
|
Result<std::unique_ptr<Service>> Service::MakeTemporaryOneshotService(
|
|
|
|
const std::vector<std::string>& args) {
|
2015-07-31 21:45:25 +02:00
|
|
|
// Parse the arguments: exec [SECLABEL [UID [GID]*] --] COMMAND ARGS...
|
|
|
|
// SECLABEL can be a - to denote default
|
|
|
|
std::size_t command_arg = 1;
|
|
|
|
for (std::size_t i = 1; i < args.size(); ++i) {
|
|
|
|
if (args[i] == "--") {
|
|
|
|
command_arg = i + 1;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (command_arg > 4 + NR_SVC_SUPP_GIDS) {
|
2019-07-30 18:34:41 +02:00
|
|
|
return Error() << "exec called with too many supplementary group ids";
|
2015-07-31 21:45:25 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
if (command_arg >= args.size()) {
|
2019-07-30 18:34:41 +02:00
|
|
|
return Error() << "exec called without command";
|
2015-07-31 21:45:25 +02:00
|
|
|
}
|
|
|
|
std::vector<std::string> str_args(args.begin() + command_arg, args.end());
|
|
|
|
|
2017-07-28 23:48:41 +02:00
|
|
|
static size_t exec_count = 0;
|
|
|
|
exec_count++;
|
|
|
|
std::string name = "exec " + std::to_string(exec_count) + " (" + Join(str_args, " ") + ")";
|
2017-04-26 02:31:06 +02:00
|
|
|
|
2017-07-28 23:48:41 +02:00
|
|
|
unsigned flags = SVC_ONESHOT | SVC_TEMPORARY;
|
2016-04-22 00:35:09 +02:00
|
|
|
unsigned namespace_flags = 0;
|
2015-07-31 21:45:25 +02:00
|
|
|
|
|
|
|
std::string seclabel = "";
|
|
|
|
if (command_arg > 2 && args[1] != "-") {
|
|
|
|
seclabel = args[1];
|
|
|
|
}
|
2017-08-03 21:54:07 +02:00
|
|
|
Result<uid_t> uid = 0;
|
2015-07-31 21:45:25 +02:00
|
|
|
if (command_arg > 3) {
|
2017-08-03 21:54:07 +02:00
|
|
|
uid = DecodeUid(args[2]);
|
2020-02-05 19:49:33 +01:00
|
|
|
if (!uid.ok()) {
|
2019-07-30 18:34:41 +02:00
|
|
|
return Error() << "Unable to decode UID for '" << args[2] << "': " << uid.error();
|
2017-05-05 02:40:33 +02:00
|
|
|
}
|
2015-07-31 21:45:25 +02:00
|
|
|
}
|
2017-08-03 21:54:07 +02:00
|
|
|
Result<gid_t> gid = 0;
|
2015-07-31 21:45:25 +02:00
|
|
|
std::vector<gid_t> supp_gids;
|
|
|
|
if (command_arg > 4) {
|
2017-08-03 21:54:07 +02:00
|
|
|
gid = DecodeUid(args[3]);
|
2020-02-05 19:49:33 +01:00
|
|
|
if (!gid.ok()) {
|
2019-07-30 18:34:41 +02:00
|
|
|
return Error() << "Unable to decode GID for '" << args[3] << "': " << gid.error();
|
2017-05-05 02:40:33 +02:00
|
|
|
}
|
2015-07-31 21:45:25 +02:00
|
|
|
std::size_t nr_supp_gids = command_arg - 1 /* -- */ - 4 /* exec SECLABEL UID GID */;
|
|
|
|
for (size_t i = 0; i < nr_supp_gids; ++i) {
|
2017-08-03 21:54:07 +02:00
|
|
|
auto supp_gid = DecodeUid(args[4 + i]);
|
2020-02-05 19:49:33 +01:00
|
|
|
if (!supp_gid.ok()) {
|
2019-07-30 18:34:41 +02:00
|
|
|
return Error() << "Unable to decode GID for '" << args[4 + i]
|
|
|
|
<< "': " << supp_gid.error();
|
2017-05-05 02:40:33 +02:00
|
|
|
}
|
2017-08-03 21:54:07 +02:00
|
|
|
supp_gids.push_back(*supp_gid);
|
2015-07-31 21:45:25 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2019-02-06 19:45:56 +01:00
|
|
|
return std::make_unique<Service>(name, flags, *uid, *gid, supp_gids, namespace_flags, seclabel,
|
2019-12-05 13:35:19 +01:00
|
|
|
nullptr, str_args, false);
|
2015-07-31 21:45:25 +02:00
|
|
|
}
|
|
|
|
|
2017-06-22 21:53:17 +02:00
|
|
|
} // namespace init
|
|
|
|
} // namespace android
|