Very Computer

   Can somebody tell me the differences between
fork and fork1 on solaris 2.6 in a multi threaded
app using pthreads. The fork/fork1 man page is not
very clear.

regards
Ranganath

Sent via Deja.com http://www.deja.com/
Share what you know. Learn what you don't.

   I thought that I should give some more info on why I need this info.
We have a MT app(pthreads). One of the threads does a fork1. The child
process does a malloc before it does an exec. Sometimes the the call to
fork1 does not return. So at this point we have two of our processes
running. If I attach the child process through dbx and check I find that
it is stuck in malloc. I have included below the relavent lines of the
trace. The child is stuck when it is trying to acquire a lock.

   Going through the man pages of fork/fork1 and pthread_atfork I
concluded that all the libraries that use locks need to call
pthread_atfork in their .init section. These man pages say that this
needs to be done to prevent any deadlocks in the child. I can probably
assume that the library supplying malloc (libc ?) is doing this
correctly.

   Earlier our code would actually use fork and not fork1, but we faced
some other probs with this and found that using fork1 solved our
problem. Now we are having the above problem using fork1. From the man
page I do not find any difference between fork & fork1 in a MT app using
pthreads.

   So can anybody tell me the differences between the two calls in such
an app. I cannot post any code as its too big and I have not tried to
simulate the problem in a sample app.

regards
Ranga

   [1] _lwp_sema_wait(0xee305e80, 0x1, 0x0, 0x0, 0xff00, 0xff), at
0xee9396f0
  [2] _park(0xee305de0, 0xee305e80, 0x0, 0xee305e5c, 0xee305e58,
0xee305e54), at 0xef24776c
  [3] _mutex_adaptive_lock(0xee9a6600, 0x4c00, 0xef265250, 0x1, 0x4d58,
0xfffeffff), at 0xef2487b0
  [4] _mutex_lock(0xee9a6600, 0xef265250, 0x0, 0x0, 0x0, 0x0), at
0xef2484f8
  [5] malloc(0x15, 0x0, 0x0, 0x0, 0x0, 0x0), at 0xee945a40

regards
Ranganath

So for an app using pthreads, there is no difference.

Roger Faulkner

     The  fork()  function  duplicates  all  the   threads   (see
     thr_create(3T))  and LWPs in the parent process in the child
     process. The fork1() function duplicates  only  the  calling
     thread (LWP) in the child process.

  POSIX Threads
     The following are the fork() semantics in programs that  use
     the  POSIX  threads  API rather than the Solaris threads API
     (programs linked with -lpthread, whether or not linked  with
     -lthread):

     The call to fork() is like a call to  fork1(), which  repli-
     cates only the calling thread. There is no call that forks a
     child with all threads and LWPs duplicated in the child.

     Note that if a program is  linked  with  both  libraries  (-
     lthread  and  -lpthread),  the POSIX semantic of fork() pre-
     vails.

With fork1() you're left with on thread; when linking w/ -lpthread,
there's no difference between fork() and fork1().

Casper
--
Expressed in this posting are my opinions.  They are in no way related
to opinions held by my employer, Sun Microsystems.
Statements on Sun products included here are not gospel and may
be fiction rather than truth.

Quote:>   Going through the man pages of fork/fork1 and pthread_atfork I
>concluded that all the libraries that use locks need to call
>pthread_atfork in their .init section. These man pages say that this
>needs to be done to prevent any deadlocks in the child. I can probably
>assume that the library supplying malloc (libc ?) is doing this
>correctly.

(All system calls and libthread calls are fork1() safe)

Casper
--
Expressed in this posting are my opinions.  They are in no way related
to opinions held by my employer, Sun Microsystems.
Statements on Sun products included here are not gospel and may
be fiction rather than truth.

A streams/device driver consultant/guru.

Please contact via email.

Randy