Is APUE wrong about linux threads description? - c

In Figure 11.2 APUE 2nd, there is a code demoing usage for threads API, as below:
#include <pthread.h>
#include <stdio.h>
pthread_t ntid;
void printids(const char *s)
{
pid_t pid;
pthread_t tid;
pid = getpid();
tid = pthread_self();
printf("%s pid %u tid %u (0x%x)\n", s,
(unsigned int)pid, (unsigned int)tid, (unsigned int)(tid));
}
void *thr_fn(void *arg)
{
printids("new thread: ");
return (void*)0;
}
int main(void)
{
int err;
err = pthread_create(&ntid, NULL, thr_fn, NULL);
if (err != 0)
return -1;
printids("main thread: ");
sleep(1);
return 0;
}
and the book says the output is like this,
$./a.out
new thread: pid 6628 ...
main thread: pid 6626 ...
Pid is different ! This is because "Linux use the clone() to implement threads, same as fork(), so the system consider the threads as separate processes who share resources".
But When I tested, I found the result is different from the APUE result, which is
$ ./a.out
main thread: pid 13301 tid 3078153920 (0xb778e6c0)
new thread: pid 13301 tid 3078151024 (0xb778db70)
the pid is the same ! so is APUE outdated ? But linux do use clone to implement threads, and in linux kernel, they are viewed as different processes. How come the process ID is the same ?

There's a high chance that your edition APUE might be outdated in the sense that it refers to LinuxThreads instead of NPTL.
Here's the relevant section from the clone(2) manpage that sheds light on this issue:
"Thread groups were a feature added in Linux 2.4 to support the POSIX threads notion of a set of threads that share a single PID. Internally, this shared PID is the so-called thread group identifier (TGID) for the thread group. Since Linux 2.4, calls to getpid(2) return the TGID of the caller.
The threads within a group can be distinguished by their (system-wide) unique thread IDs (TID). A new thread's TID is available as the function result returned to the caller of clone(), and a thread can obtain its own TID using gettid(2)."

Related

segfault with clone() and printf

I'm trying to experiment with how clone() is implemented for threads in Linux 3.10.0-327.3.1.el7.x86_64
I'm running this piece of code and having occasional segfaults. I know if I use CLONE_THREAD then there's no way to check if the thread finished, but why does printf cause problems? How does the Pthread library handle this issue? Without the printfs there's no segfault.
#define STACK_SIZE (1ULL<<22) //4MB
int tmp = 10;
int threadfunc(void *ptr)
{
printf("i'm here\n");
tmp++;
return 0;
}
int main()
{
char *child_stack = malloc(STACK_SIZE);
int child_pid = clone(&threadfunc, child_stack+(STACK_SIZE)-1, CLONE_THREAD|CLONE_SIGHAND|CLONE_VM|CLONE_FS|CLONE_FILES|SIGCHLD, NULL);
printf("my pid: %d\n", getpid());
printf("child pid: %d\n", child_pid);
printf("tmp is: %d\n", val);
return 0;
}
The printf function doesn't understand that it's being called in a clone
task; it is designed around the glibc pthread library.
What you're doing is similar to interrupting a printf with an asynchronous signal, and re-entering printf from the signal.
How the library handles the issue is that it uses clone as part of a more complicated implementation of threads, which "dresses up" the cloned task with a thread identity. This thread identity allows printf to lock a mutex around the stream operation.
What is the thread identity? What it means is that a thread-related function in the C library can inquire, in some way, "what thread is calling me?" to obtain an ID, which maps to a thread descriptor structure. The clone function alone won't set up this identity.

process id of threads in the same process

The following code is for printing the process id of the 2 threads linux(ubuntu 14.04)
#include<pthread.h>
#include<stdio.h>
#include <unistd.h>
void* thread_function (void* arg)
{
fprintf (stderr, "child thread pid is %d\n", (int) getpid ());
/* Spin forever. */
while (1);
return NULL;
}
int main ()
{
pthread_t thread;
fprintf (stderr, "main thread pid is %d\n", (int) getpid ());
pthread_create (&thread, NULL, &thread_function, NULL);
/* Spin forever. */
while (1);
return 0;
}
And the output is
main thread pid is 3614
child thread pid is 3614
But shouldn't be the process id be different since GNU/Linux,threads are implemented as processes??
There's a terminology conflict here. It's true that each thread is a separate process as far as the Linux kernel is concerned. And therefore Linux assigns a new PID to each thread.
But that's not how POSIX works: according to POSIX all threads in a process should share the same PID. The Linux kernel calls this "Thread Group IDs" (TGID), and the getpid() function actually returns the TGID, in order to be POSIX-compliant.
Three separate concepts: the process id (getpid), the pthreads thread id (pthread_self) and the underlying linux thread id (gettid).
There is no glibc wrapper for gettid so it amounts to this
pid_t gettid(void)
{
return(syscall(SYS_gettid));
}
On the common pthreads programming level you shouldn't be concerned how the threads are implemented (though pdw explained it well enough). All this stuff is made deliberately opaque. There is no use for gettid in any pthreads function. They all require the pthreads thread id.
There are two main reasons why people ask about linux thread ids. One, they want to understand the relationship of linux thread ids in relation to some system utility e.g. ps, htop. Two, there are literally a handful of linux specific system calls in which the linux tid is useful.

How to get thread id of a pthread in linux c program?

In a Linux C program, how do I print the thread id of a thread created by the pthread library? For example like how we can get pid of a process by getpid().
What? The person asked for Linux specific, and the equivalent of getpid(). Not BSD or Apple. The answer is gettid() and returns an integral type. You will have to call it using syscall(), like this:
#include <sys/types.h>
#include <unistd.h>
#include <sys/syscall.h>
....
pid_t x = syscall(__NR_gettid);
While this may not be portable to non-linux systems, the threadid is directly comparable and very fast to acquire. It can be printed (such as for LOGs) like a normal integer.
pthread_self() function will give the thread id of current thread.
pthread_t pthread_self(void);
The pthread_self() function returns the Pthread handle of the calling thread. The pthread_self() function does NOT return the integral thread of the calling thread. You must use pthread_getthreadid_np() to return an integral identifier for the thread.
NOTE:
pthread_id_np_t tid;
tid = pthread_getthreadid_np();
is significantly faster than these calls, but provides the same behavior.
pthread_id_np_t tid;
pthread_t self;
self = pthread_self();
pthread_getunique_np(&self, &tid);
As noted in other answers, pthreads does not define a platform-independent way to retrieve an integral thread ID.
On Linux systems, you can get thread ID thus:
#include <sys/types.h>
pid_t tid = gettid();
On many BSD-based platforms, this answer https://stackoverflow.com/a/21206357/316487 gives a non-portable way.
However, if the reason you think you need a thread ID is to know whether you're running on the same or different thread to another thread you control, you might find some utility in this approach
static pthread_t threadA;
// On thread A...
threadA = pthread_self();
// On thread B...
pthread_t threadB = pthread_self();
if (pthread_equal(threadA, threadB)) printf("Thread B is same as thread A.\n");
else printf("Thread B is NOT same as thread A.\n");
If you just need to know if you're on the main thread, there are additional ways, documented in answers to this question how can I tell if pthread_self is the main (first) thread in the process?.
pid_t tid = syscall(SYS_gettid);
Linux provides such system call to allow you get id of a thread.
You can use pthread_self()
The parent gets to know the thread id after the pthread_create() is executed sucessfully, but while executing the thread if we want to access the thread id we have to use the function pthread_self().
This single line gives you pid , each threadid and spid.
printf("before calling pthread_create getpid: %d getpthread_self: %lu tid:%lu\n",getpid(), pthread_self(), syscall(SYS_gettid));
I think not only is the question not clear but most people also are not cognizant of the difference. Examine the following saying,
POSIX thread IDs are not the same as the thread IDs returned by the
Linux specific gettid() system call. POSIX thread IDs are assigned
and maintained by the threading implementation. The thread ID returned
by gettid() is a number (similar to a process ID) that is assigned by
the kernel. Although each POSIX thread has a unique kernel thread ID
in the Linux NPTL threading implementation, an application generally
doesn’t need to know about the kernel IDs (and won’t be portable if it
depends on knowing them).
Excerpted from: The Linux Programming Interface: A Linux and UNIX System Programming Handbook, Michael Kerrisk
IMHO, there is only one portable way that pass a structure in which define a variable holding numbers in an ascending manner e.g. 1,2,3... to per thread. By doing this, threads' id can be kept track. Nonetheless, int pthread_equal(tid1, tid2) function should be used.
if (pthread_equal(tid1, tid2)) printf("Thread 2 is same as thread 1.\n");
else printf("Thread 2 is NOT same as thread 1.\n");
pthread_getthreadid_np wasn't on my Mac os x. pthread_t is an opaque type. Don't beat your head over it. Just assign it to void* and call it good. If you need to printf use %p.
There is also another way of getting thread id. While creating threads with
int pthread_create(pthread_t * thread, const pthread_attr_t * attr, void * (*start_routine)(void *), void *arg);
function call; the first parameter pthread_t * thread is actually a thread id (that is an unsigned long int defined in bits/pthreadtypes.h). Also, the last argument void *arg is the argument that is passed to void * (*start_routine) function to be threaded.
You can create a structure to pass multiple arguments and send a pointer to a structure.
typedef struct thread_info {
pthread_t thread;
//...
} thread_info;
//...
tinfo = malloc(sizeof(thread_info) * NUMBER_OF_THREADS);
//...
pthread_create (&tinfo[i].thread, NULL, handler, (void*)&tinfo[i]);
//...
void *handler(void *targs) {
thread_info *tinfo = targs;
// here you get the thread id with tinfo->thread
}
For different OS there is different answer. I find a helper here.
You can try this:
#include <unistd.h>
#include <sys/syscall.h>
int get_thread_id() {
#if defined(__linux__)
return syscall(SYS_gettid);
#elif defined(__FreeBSD__)
long tid;
thr_self(&tid);
return (int)tid;
#elif defined(__NetBSD__)
return _lwp_self();
#elif defined(__OpenBSD__)
return getthrid();
#else
return getpid();
#endif
}
Platform-independent way (starting from c++11) is:
#include <thread>
std::this_thread::get_id();
You can also write in this manner and it does the same. For eg:
for(int i=0;i < total; i++)
{
pthread_join(pth[i],NULL);
cout << "SUM of thread id " << pth[i] << " is " << args[i].sum << endl;
}
This program sets up an array of pthread_t and calculate sum on each. So it is printing the sum of each thread with thread id.

pthread_create from a child process

I'm sure I'm missing something basic, but I'm writing a program that fork()s several child processes, each of which create several pthreads. It seems that the pthread_create call never works from a child process. Here is sample code to explain what I mean:
#include <stdio.h>
#include <unistd.h>
#include <stdlib.h>
#include <pthread.h>
// Listen for a request from user proc i
void *wait_for_req(void *cvoid) {
int req;
int read_ret;
printf("new thread started\n");
pthread_exit(NULL);
}
void spawn_read_threads(int proc_num, int n) {
int i;
printf("About to spawn %d read threads\n", n);
for (i=0; i<n; i++) {
pthread_t *t;
printf("spawning new thread\n");
int create_result = pthread_create(t, NULL, wait_for_req, NULL);
printf("_create returned %d\n", create_result);
pthread_join(*t, NULL);
}
}
int main() {
if (!fork())
spawn_read_threads(0, 1);
}
The output of this program is
About to spawn 1 read threads
spawning new thread
But if I comment out if comment out if (!fork()):
About to spawn 1 read threads
spawning new thread
_create returned 0
new thread started
So why doesn't execution get through create_result in the first case?
And if this is useful:
rob#ubuntu:/mnt/hgfs/Virtual Machines$ uname -a
Linux ubuntu 2.6.32-24-generic #42-Ubuntu SMP Fri Aug 20 14:24:04 UTC 2010 i686 GNU/Linux
I can see one immediate problem with your code: pthread_create takes the address of an existing pthread_t variable and stores the new thread ID in it. You're passing an uninitialized pointer instead. You need to do something like:
pthread_t tid;
int create_result = pthread_create(&tid, NULL, wait_for_req, NULL);
I'd say you were just lucky when it worked once, because your current code essentially causes undefined behaviour.
Besides the errors you found in your answer, you are not supposed to call pthread_* from within the forked process. After, fork() only async-signal safe functions should be used. You must exec() first, and then you may create some pthreads.
pthread_t t;
printf("spawning new thread\n");
int create_result = pthread_create(&t, NULL, wait_for_req, NULL);
printf("_create returned %d\n", create_result);
pthread_join(t, NULL);

How does POSIX Threads work in linux?

I thought pthread uses clone to spawn one new thread in linux. But if so, all of the threads should have their seperate pid. Otherwise, if they have the same pid, the global variables in the libc seem to be shared. However, as I ran the following program, I got the same pid but the different address of errno.
extern errno;
void*
f(void *arg)
{
printf("%u,%p\n", getpid(), &errno);
fflush(stdin);
return NULL;
}
int
main(int argc, char **argv)
{
pthread_t tid;
pthread_create(&tid, NULL, f, NULL);
printf("%u,%p\n", getpid(), &errno);
fflush(stdin);
pthread_join(tid, NULL);
return 0;
}
Then, why?
I'm not sure exactly how clone() is used when pthread_create() is called. That said, looking at the clone() man page, it looks like there is a flag called CLONE_THREAD which:
If CLONE_THREAD is set, the child is
placed in the same thread group as the
calling process. To make the remainder
of the discussion of CLONE_THREAD more
readable, the term "thread" is used to
refer to the processes within a thread
group.
Thread groups were a feature added in
Linux 2.4 to support the POSIX threads
notion of a set of threads that share
a single PID. Internally, this shared
PID is the so-called thread group
identifier (TGID) for the thread
group. Since Linux 2.4, calls to
getpid(2) return the TGID of the
caller.
It then goes on to talk about a gettid() function for getting the unique ID of an individual thread within a process. Modifying your code:
#include <stdio.h>
#include <pthread.h>
#include <sys/types.h>
#include <sys/syscall.h>
#include <unistd.h>
int errno;
void*
f(void *arg)
{
printf("%u,%p, %u\n", getpid(), &errno, syscall(SYS_gettid));
fflush(stdin);
return NULL;
}
int
main(int argc, char **argv)
{
pthread_t tid;
pthread_create(&tid, NULL, f, NULL);
printf("%u,%p, %u\n", getpid(), &errno, syscall(SYS_gettid));
fflush(stdin);
pthread_join(tid, NULL);
return 0;
}
(make sure to use "-lpthread"!) we can see that the individual thread id is indeed unique, while the pid remains the same.
rascher#coltrane:~$ ./a.out
4109,0x804a034, 4109
4109,0x804a034, 4110
Global variables: your mistake is that errno is not a global variable but a macro that expands to an lvalue of type int. In practice, it expands to (*__errno_location()) or similar.
getpid is a library function that returns the process id in the POSIX sense of process, not the bogus Linux per-clone pid. Nowadays Linux has the minimal kernel-level functionality necessary to make near-POSIX-compliance possible with respect to threads, but most of it still depends on ugly hacks at the userspace libc level.

Resources