ClickHouse/src/Common/ShellCommand.cpp

234 lines
8.0 KiB
C++
Raw Normal View History

2015-12-13 08:51:28 +00:00
#include <sys/types.h>
#include <sys/wait.h>
#include <fcntl.h>
#include <dlfcn.h>
#include <Common/Exception.h>
#include <Common/ShellCommand.h>
#include <Common/PipeFDs.h>
#include <common/logger_useful.h>
#include <common/errnoToString.h>
#include <IO/WriteHelpers.h>
2020-02-17 14:27:09 +00:00
#include <unistd.h>
#include <csignal>
2015-12-13 08:51:28 +00:00
2019-07-05 13:48:47 +00:00
namespace
{
/// By these return codes from the child process, we learn (for sure) about errors when creating it.
enum class ReturnCodes : int
{
CANNOT_DUP_STDIN = 0x55555555, /// The value is not important, but it is chosen so that it's rare to conflict with the program return code.
CANNOT_DUP_STDOUT = 0x55555556,
CANNOT_DUP_STDERR = 0x55555557,
CANNOT_EXEC = 0x55555558,
};
}
2015-12-13 08:51:28 +00:00
namespace DB
{
2019-02-03 09:57:12 +00:00
namespace ErrorCodes
2015-12-13 08:51:28 +00:00
{
2019-02-03 21:30:45 +00:00
extern const int CANNOT_DLSYM;
extern const int CANNOT_FORK;
extern const int CANNOT_WAITPID;
extern const int CHILD_WAS_NOT_EXITED_NORMALLY;
extern const int CANNOT_CREATE_CHILD_PROCESS;
2015-12-13 08:51:28 +00:00
}
2019-08-03 11:02:40 +00:00
ShellCommand::ShellCommand(pid_t pid_, int in_fd_, int out_fd_, int err_fd_, bool terminate_in_destructor_)
: pid(pid_)
, terminate_in_destructor(terminate_in_destructor_)
2019-08-03 11:02:40 +00:00
, in(in_fd_)
, out(out_fd_)
, err(err_fd_) {}
Poco::Logger * ShellCommand::getLogger()
{
return &Poco::Logger::get("ShellCommand");
}
ShellCommand::~ShellCommand()
{
if (terminate_in_destructor)
{
2020-05-23 22:24:01 +00:00
LOG_TRACE(getLogger(), "Will kill shell command pid {} with SIGTERM", pid);
int retcode = kill(pid, SIGTERM);
if (retcode != 0)
2020-05-23 22:24:01 +00:00
LOG_WARNING(getLogger(), "Cannot kill shell command pid {} errno '{}'", pid, errnoToString(retcode));
}
else if (!wait_called)
tryWait();
}
2015-12-13 08:51:28 +00:00
void ShellCommand::logCommand(const char * filename, char * const argv[])
{
2020-05-23 20:13:34 +00:00
std::stringstream args;
2019-09-27 12:22:27 +00:00
for (int i = 0; argv != nullptr && argv[i] != nullptr; ++i)
{
if (i > 0)
2020-05-23 20:13:34 +00:00
args << ", ";
2019-12-09 14:49:21 +00:00
2019-09-27 12:22:27 +00:00
/// NOTE: No escaping is performed.
2020-05-23 20:13:34 +00:00
args << "'" << argv[i] << "'";
}
2020-05-23 22:24:01 +00:00
LOG_TRACE(ShellCommand::getLogger(), "Will start shell command '{}' with arguments {}", filename, args.str());
}
std::unique_ptr<ShellCommand> ShellCommand::executeImpl(const char * filename, char * const argv[], bool pipe_stdin_only, bool terminate_in_destructor)
2015-12-13 08:51:28 +00:00
{
logCommand(filename, argv);
/** Here it is written that with a normal call `vfork`, there is a chance of deadlock in multithreaded programs,
2020-02-13 14:25:17 +00:00
* because of the resolving of symbols in the shared library
* http://www.oracle.com/technetwork/server-storage/solaris10/subprocess-136439.html
* Therefore, separate the resolving of the symbol from the call.
*/
static void * real_vfork = dlsym(RTLD_DEFAULT, "vfork");
2015-12-13 08:51:28 +00:00
if (!real_vfork)
throwFromErrno("Cannot find symbol vfork in myself", ErrorCodes::CANNOT_DLSYM);
2015-12-13 08:51:28 +00:00
PipeFDs pipe_stdin;
PipeFDs pipe_stdout;
PipeFDs pipe_stderr;
2015-12-13 08:51:28 +00:00
pid_t pid = reinterpret_cast<pid_t(*)()>(real_vfork)();
2015-12-13 08:51:28 +00:00
if (-1 == pid)
throwFromErrno("Cannot vfork", ErrorCodes::CANNOT_FORK);
2015-12-13 08:51:28 +00:00
if (0 == pid)
{
/// We are in the freshly created process.
2015-12-13 08:51:28 +00:00
/// Why `_exit` and not `exit`? Because `exit` calls `atexit` and destructors of thread local storage.
/// And there is a lot of garbage (including, for example, mutex is blocked). And this can not be done after `vfork` - deadlock happens.
2015-12-13 08:51:28 +00:00
/// Replace the file descriptors with the ends of our pipes.
if (STDIN_FILENO != dup2(pipe_stdin.fds_rw[0], STDIN_FILENO))
_exit(int(ReturnCodes::CANNOT_DUP_STDIN));
2015-12-13 08:51:28 +00:00
if (!pipe_stdin_only)
{
if (STDOUT_FILENO != dup2(pipe_stdout.fds_rw[1], STDOUT_FILENO))
_exit(int(ReturnCodes::CANNOT_DUP_STDOUT));
2015-12-13 08:51:28 +00:00
if (STDERR_FILENO != dup2(pipe_stderr.fds_rw[1], STDERR_FILENO))
_exit(int(ReturnCodes::CANNOT_DUP_STDERR));
}
2015-12-13 08:51:28 +00:00
// Reset the signal mask: it may be non-empty and will be inherited
// by the child process, which might not expect this.
sigset_t mask;
sigemptyset(&mask);
sigprocmask(0, nullptr, &mask);
sigprocmask(SIG_UNBLOCK, &mask, nullptr);
execv(filename, argv);
/// If the process is running, then `execv` does not return here.
2015-12-13 08:51:28 +00:00
_exit(int(ReturnCodes::CANNOT_EXEC));
}
2015-12-13 08:51:28 +00:00
std::unique_ptr<ShellCommand> res(new ShellCommand(pid, pipe_stdin.fds_rw[1], pipe_stdout.fds_rw[0], pipe_stderr.fds_rw[0], terminate_in_destructor));
2015-12-13 08:51:28 +00:00
2020-05-23 22:24:01 +00:00
LOG_TRACE(getLogger(), "Started shell command '{}' with pid {}", filename, pid);
/// Now the ownership of the file descriptors is passed to the result.
pipe_stdin.fds_rw[1] = -1;
pipe_stdout.fds_rw[0] = -1;
pipe_stderr.fds_rw[0] = -1;
2015-12-13 08:51:28 +00:00
return res;
2015-12-13 08:51:28 +00:00
}
std::unique_ptr<ShellCommand> ShellCommand::execute(const std::string & command, bool pipe_stdin_only, bool terminate_in_destructor)
2015-12-13 08:51:28 +00:00
{
/// Arguments in non-constant chunks of memory (as required for `execv`).
/// Moreover, their copying must be done before calling `vfork`, so after `vfork` do a minimum of things.
std::vector<char> argv0("sh", &("sh"[3]));
std::vector<char> argv1("-c", &("-c"[3]));
std::vector<char> argv2(command.data(), command.data() + command.size() + 1);
2015-12-13 08:51:28 +00:00
char * const argv[] = { argv0.data(), argv1.data(), argv2.data(), nullptr };
2015-12-13 08:51:28 +00:00
return executeImpl("/bin/sh", argv, pipe_stdin_only, terminate_in_destructor);
2015-12-13 08:51:28 +00:00
}
std::unique_ptr<ShellCommand> ShellCommand::executeDirect(const std::string & path, const std::vector<std::string> & arguments, bool terminate_in_destructor)
2015-12-13 08:51:28 +00:00
{
size_t argv_sum_size = path.size() + 1;
for (const auto & arg : arguments)
argv_sum_size += arg.size() + 1;
2015-12-13 08:51:28 +00:00
std::vector<char *> argv(arguments.size() + 2);
std::vector<char> argv_data(argv_sum_size);
WriteBuffer writer(argv_data.data(), argv_sum_size);
2015-12-13 08:51:28 +00:00
argv[0] = writer.position();
writer.write(path.data(), path.size() + 1);
2015-12-13 08:51:28 +00:00
for (size_t i = 0, size = arguments.size(); i < size; ++i)
{
argv[i + 1] = writer.position();
writer.write(arguments[i].data(), arguments[i].size() + 1);
}
2015-12-13 08:51:28 +00:00
argv[arguments.size() + 1] = nullptr;
2015-12-13 08:51:28 +00:00
return executeImpl(path.data(), argv.data(), false, terminate_in_destructor);
2015-12-13 08:51:28 +00:00
}
int ShellCommand::tryWait()
{
wait_called = true;
2020-05-23 22:24:01 +00:00
LOG_TRACE(getLogger(), "Will wait for shell command pid {}", pid);
int status = 0;
if (-1 == waitpid(pid, &status, 0))
throwFromErrno("Cannot waitpid", ErrorCodes::CANNOT_WAITPID);
2015-12-13 08:51:28 +00:00
2020-05-23 22:24:01 +00:00
LOG_TRACE(getLogger(), "Wait for shell command pid {} completed with status {}", pid, status);
if (WIFEXITED(status))
return WEXITSTATUS(status);
2015-12-13 08:51:28 +00:00
if (WIFSIGNALED(status))
throw Exception("Child process was terminated by signal " + toString(WTERMSIG(status)), ErrorCodes::CHILD_WAS_NOT_EXITED_NORMALLY);
2015-12-13 08:51:28 +00:00
if (WIFSTOPPED(status))
throw Exception("Child process was stopped by signal " + toString(WSTOPSIG(status)), ErrorCodes::CHILD_WAS_NOT_EXITED_NORMALLY);
2015-12-13 08:51:28 +00:00
throw Exception("Child process was not exited normally by unknown reason", ErrorCodes::CHILD_WAS_NOT_EXITED_NORMALLY);
2015-12-13 08:51:28 +00:00
}
void ShellCommand::wait()
{
int retcode = tryWait();
if (retcode != EXIT_SUCCESS)
{
switch (retcode)
{
case int(ReturnCodes::CANNOT_DUP_STDIN):
throw Exception("Cannot dup2 stdin of child process", ErrorCodes::CANNOT_CREATE_CHILD_PROCESS);
case int(ReturnCodes::CANNOT_DUP_STDOUT):
throw Exception("Cannot dup2 stdout of child process", ErrorCodes::CANNOT_CREATE_CHILD_PROCESS);
case int(ReturnCodes::CANNOT_DUP_STDERR):
throw Exception("Cannot dup2 stderr of child process", ErrorCodes::CANNOT_CREATE_CHILD_PROCESS);
case int(ReturnCodes::CANNOT_EXEC):
throw Exception("Cannot execv in child process", ErrorCodes::CANNOT_CREATE_CHILD_PROCESS);
default:
throw Exception("Child process was exited with return code " + toString(retcode), ErrorCodes::CHILD_WAS_NOT_EXITED_NORMALLY);
}
}
2015-12-13 08:51:28 +00:00
}
}