/**
* Starts the agent process. May throw arbitrary exceptions.
*/
virtual pid_t start() {
this_thread::disable_interruption di;
this_thread::disable_syscall_interruption dsi;
string exeFilename = getExeFilename();
SocketPair fds;
int e, ret;
pid_t pid;
/* Create feedback fd for this agent process. We'll send some startup
* arguments to this agent process through this fd, and we'll receive
* startup information through it as well.
*/
fds = createUnixSocketPair();
pid = syscalls::fork();
if (pid == 0) {
// Child
/* Make sure file descriptor FEEDBACK_FD refers to the newly created
* feedback fd (fds[1]) and close all other file descriptors.
* In this child process we don't care about the original FEEDBACK_FD
* (which is the Watchdog's communication channel to the agents starter.)
*
* fds[1] is guaranteed to be != FEEDBACK_FD because the watchdog
* is started with FEEDBACK_FD already assigned.
*/
syscalls::close(fds[0]);
if (syscalls::dup2(fds[1], FEEDBACK_FD) == -1) {
/* Something went wrong, report error through feedback fd. */
e = errno;
try {
writeArrayMessage(fds[1],
"system error before exec",
"dup2() failed",
toString(e).c_str(),
NULL);
_exit(1);
} catch (...) {
fprintf(stderr, "Passenger Watchdog: dup2() failed: %s (%d)\n",
strerror(e), e);
fflush(stderr);
_exit(1);
}
}
closeAllFileDescriptors(FEEDBACK_FD);
/* Become the process group leader so that the watchdog can kill the
* agent as well as all its descendant processes. */
setpgid(getpid(), getpid());
setOomScore(oldOomScore);
try {
execProgram();
} catch (...) {
fprintf(stderr, "PassengerWatchdog: execProgram() threw an exception\n");
fflush(stderr);
_exit(1);
}
e = errno;
try {
writeArrayMessage(FEEDBACK_FD,
"exec error",
toString(e).c_str(),
NULL);
} catch (...) {
fprintf(stderr, "Passenger Watchdog: could not execute %s: %s (%d)\n",
exeFilename.c_str(), strerror(e), e);
fflush(stderr);
}
_exit(1);
} else if (pid == -1) {
// Error
e = errno;
throw SystemException("Cannot fork a new process", e);
} else {
// Parent
FileDescriptor feedbackFd = fds[0];
vector<string> args;
fds[1].close();
this_thread::restore_interruption ri(di);
this_thread::restore_syscall_interruption rsi(dsi);
ScopeGuard failGuard(boost::bind(killAndWait, pid));
/* Send startup arguments. Ignore EPIPE and ECONNRESET here
* because the child process might have sent an feedback message
* without reading startup arguments.
*/
try {
sendStartupArguments(pid, feedbackFd);
} catch (const SystemException &ex) {
if (ex.code() != EPIPE && ex.code() != ECONNRESET) {
throw SystemException(string("Unable to start the ") + name() +
": an error occurred while sending startup arguments",
ex.code());
}
}
// Now read its feedback.
try {
ret = readArrayMessage(feedbackFd, args);
} catch (const SystemException &e) {
if (e.code() == ECONNRESET) {
ret = false;
} else {
throw SystemException(string("Unable to start the ") + name() +
": unable to read its startup information",
e.code());
}
}
if (!ret) {
this_thread::disable_interruption di2;
this_thread::disable_syscall_interruption dsi2;
int status;
/* The feedback fd was prematurely closed for an unknown reason.
* Did the agent process crash?
*
* We use timedWaitPid() here because if the process crashed
* because of an uncaught exception, the file descriptor
* might be closed before the process has printed an error
* message, so we give it some time to print the error
* before we kill it.
*/
ret = timedWaitPid(pid, &status, 5000);
if (ret == 0) {
/* Doesn't look like it; it seems it's still running.
* We can't do anything without proper feedback so kill
* the agent process and throw an exception.
*/
failGuard.runNow();
throw RuntimeException(string("Unable to start the ") + name() +
": it froze and reported an unknown error during its startup");
} else if (ret != -1 && WIFSIGNALED(status)) {
/* Looks like a crash which caused a signal. */
throw RuntimeException(string("Unable to start the ") + name() +
": it seems to have been killed with signal " +
getSignalName(WTERMSIG(status)) + " during startup");
} else if (ret == -1) {
/* Looks like it exited after detecting an error. */
throw RuntimeException(string("Unable to start the ") + name() +
": it seems to have crashed during startup for an unknown reason");
} else {
/* Looks like it exited after detecting an error, but has an exit code. */
throw RuntimeException(string("Unable to start the ") + name() +
": it seems to have crashed during startup for an unknown reason, "
"with exit code " + toString(WEXITSTATUS(status)));
}
}
if (args[0] == "system error before exec") {
throw SystemException(string("Unable to start the ") + name() +
": " + args[1], atoi(args[2]));
} else if (args[0] == "exec error") {
e = atoi(args[1]);
if (e == ENOENT) {
throw RuntimeException(string("Unable to start the ") + name() +
" because its executable (" + getExeFilename() + ") "
"doesn't exist. This probably means that your "
"Phusion Passenger installation is broken or "
"incomplete. Please reinstall Phusion Passenger");
} else {
throw SystemException(string("Unable to start the ") + name() +
" because exec(\"" + getExeFilename() + "\") failed",
atoi(args[1]));
}
} else if (!processStartupInfo(pid, feedbackFd, args)) {
throw RuntimeException(string("The ") + name() +
" sent an unknown startup info message '" +
args[0] + "'");
}
lock_guard<boost::mutex> l(lock);
this->feedbackFd = feedbackFd;
this->pid = pid;
failGuard.clear();
return pid;
}
}
void
threadMain() {
while (!this_thread::interruption_requested()) {
syscalls::sleep(60 * 60);
begin_touch:
this_thread::disable_interruption di;
this_thread::disable_syscall_interruption dsi;
// Fork a process which touches everything in the server instance dir.
pid_t pid = syscalls::fork();
if (pid == 0) {
// Child
int prio, ret, e;
closeAllFileDescriptors(2);
// Make process nicer.
do {
prio = getpriority(PRIO_PROCESS, getpid());
} while (prio == -1 && errno == EINTR);
if (prio != -1) {
prio++;
if (prio > 20) {
prio = 20;
}
do {
ret = setpriority(PRIO_PROCESS, getpid(), prio);
} while (ret == -1 && errno == EINTR);
} else {
perror("getpriority");
}
do {
ret = chdir(wo->instanceDir->getPath().c_str());
} while (ret == -1 && errno == EINTR);
if (ret == -1) {
e = errno;
fprintf(stderr, "chdir(\"%s\") failed: %s (%d)\n",
wo->instanceDir->getPath().c_str(),
strerror(e), e);
fflush(stderr);
_exit(1);
}
setOomScore(oldOomScore);
execlp("/bin/sh", "/bin/sh", "-c", "find . | xargs touch", (char *) 0);
e = errno;
fprintf(stderr, "Cannot execute 'find . | xargs touch': %s (%d)\n",
strerror(e), e);
fflush(stderr);
_exit(1);
} else if (pid == -1) {
// Error
P_WARN("Could not touch the server instance directory because "
"fork() failed. Retrying in 2 minutes...");
this_thread::restore_interruption si(di);
this_thread::restore_syscall_interruption rsi(dsi);
syscalls::sleep(60 * 2);
goto begin_touch;
} else {
syscalls::waitpid(pid, NULL, 0);
}
}
}
