DevOps

Wednesday, 13 October 2010

fork command

System Calls                                              fork(2)

NAME
     fork, fork1, forkall - create a new process

SYNOPSIS
     #include <sys/types.h>
     #include <unistd.h>

     pid_t fork(void);

     pid_t fork1(void);

     pid_t forkall(void);

DESCRIPTION
     The fork(), fork1(), and forkall() functions  create  a  new
     process.  The  address  space of the new process (child pro-
     cess) is an exact copy of the address space of  the  calling
     process  (parent  process).  The  child process inherits the
     following attributes from the parent process:

       o  real user ID, real group ID, effective user ID,  effec-
          tive group ID

       o  environment

       o  open file descriptors

       o  close-on-exec flags (see exec(2))

       o  signal handling settings (that  is,  SIG_DFL,  SIG_IGN,
          SIG_HOLD, function address)

       o  supplementary group IDs

       o  set-user-ID mode bit

       o  set-group-ID mode bit

       o  profiling on/off status

       o  nice value (see  nice(2))

       o  scheduler class (see priocntl(2))

       o  all attached shared memory segments (see shmop(2))

       o  process group ID -- memory mappings (see mmap(2))

       o  session ID (see exit(2))

       o  current working directory

SunOS 5.10          Last change: 19 Jul 2004                    1

System Calls                                              fork(2)

       o  root directory

       o  file mode creation mask (see umask(2))

       o  resource limits (see  getrlimit(2))

       o  controlling terminal

       o  saved user ID and group ID

       o  task ID and project ID

       o  processor bindings (see processor_bind(2))

       o  processor set bindings (see pset_bind(2))

       o  process privilege sets (see getppriv(2))

       o  process flags (see getpflags(2))

       o  active contract templates (see contract(4))

     Scheduling priority and any per-process  scheduling  parame-
     ters  that are specific to a given scheduling class might or
     might not be inherited according to the policy of that  par-
     ticular  class (see priocntl(2)). The child process might or
     might not be in the same process contract as the parent (see
     process(4)).  The child process differs from the parent pro-
     cess in the following ways:

       o  The child process has a unique process  ID  which  does
          not match any active process group ID.

       o  The child process has a  different  parent  process  ID
          (that is, the process ID of the parent process).

       o  The child process has its own copy of the parent's file
          descriptors  and directory streams. Each of the child's
          file descriptors shares a common file pointer with  the
          corresponding file descriptor of the parent.

       o  Each shared memory segment  remains  attached  and  the
          value of shm_nattach is incremented by 1.

       o  All semadj values are cleared (see semop(2)).

       o  Process locks, text locks, data locks, and other memory
          locks are not inherited by the child (see plock(3C) and
          memcntl(2)).

       o

SunOS 5.10          Last change: 19 Jul 2004                    2

System Calls                                              fork(2)

          The  child  process's   tms   structure   is   cleared:
          tms_utime,  stime, cutime, and cstime are set to 0 (see
          times(2)).

       o  The child processes resource utilizations are set to 0;
          see  getrlimit(2).  The it_value and it_interval values
          for the ITIMER_REAL timer are reset to 0;  see  getiti-
          mer(2).

       o  The set of signals pending for  the  child  process  is
          initialized to the empty set.

       o  Timers created by timer_create(3RT) are  not  inherited
          by the child process.

       o  No asynchronous input or asynchronous output operations
          are inherited by the child.

       o  Any preferred hardware address tranlsation  sizes  (see
          memcntl(2)) are inherited by the child.

       o  The child process holds no contracts (see contract(4)).

     Record locks set by the parent process are not inherited  by
     the child process (see fcntl(2)).

     Although any open door descriptors in the parent are  shared
     by the child, only the parent will receive a door invocation
     from clients even if the door  descriptor  is  open  in  the
     child.  If a descriptor is closed in the parent, attempts to
     operate on the door descriptor will fail even if it is still
     open in the child.

  Threads
     A call to forkall() replicates in the child process  all  of
     the  threads  (see thr_create(3C) and pthread_create(3C)) in
     the parent process. A call to fork1()  replicates  only  the
     calling thread in the child process.

     In Solaris 10, a call to fork() is identical to  a  call  to
     fork1();  only the calling thread is replicated in the child
     process. This is the POSIX-specified behavior for fork().

     In previous releases of  Solaris,  the  behavior  of  fork()
     depended  on  whether or not the application was linked with
     the  POSIX  threads  library.  When  linked  with   -lthread
     (Solaris  Threads)  but  not  linked  with  -lpthread (POSIX
     Threads), fork() was the same  as  forkall().   When  linked
     with  -lpthread,  whether  or not also linked with -lthread,
     fork() was the same as fork1().

SunOS 5.10          Last change: 19 Jul 2004                    3

System Calls                                              fork(2)

     In Solaris 10, neither -lthread nor  -lpthread  is  required
     for  multithreaded applications. The standard C library pro-
     vides all threading support for  both  sets  of  application
     programming    interfaces.     Applications   that   require
     replicate-all fork semantics must call forkall().

  fork() Safety
     If a multithreaded application calls fork() or fork1(),  and
     the  child  does  more  than  simply call one of the exec(2)
     functions, there is a possibility of deadlock  occurring  in
     the  child. The application should use pthread_atfork(3C) to
     ensure safety with respect to this deadlock. Should there be
     any outstanding mutexes throughout the process, the applica-
     tion should call pthread_atfork() to wait  for  and  acquire
     those  mutexes prior to calling fork() or fork1(). See  "MT-
     Level of Libraries" on the attributes(5) manual page.

RETURN VALUES
     Upon successful completion, fork(), fork1(),  and  forkall()
     return  0  to the child process and return the process ID of
     the  child  process  to  the  parent   process.   Otherwise,
     (pid_t)-1  is  returned to the parent process, no child pro-
     cess is created, and errno is set to indicate the error.

ERRORS
     The fork(), fork1(), and forkall()function will fail if:

     EAGAIN          A resource control or  limit  on  the  total
                     number  of  processes,  tasks  or LWPs under
                     execution by a single user,  task,  project,
                     or  zone  has  been  exceeded,  or the total
                     amount of system memory  available  is  tem-
                     porarily insufficient to duplicate this pro-
                     cess.

     ENOMEM          There is not enough swap space.

     EPERM           The  {PRIV_PROC_FORK}   privilege   is   not
                     asserted in the effective set of the calling
                     process.

ATTRIBUTES
     See attributes(5) for descriptions of the  following  attri-
     butes:

SunOS 5.10          Last change: 19 Jul 2004                    4

System Calls                                              fork(2)

     ____________________________________________________________
   |        ATTRIBUTE TYPE       |        ATTRIBUTE VALUE      |
   | ____________________________|_____________________________|_
   |  Interface Stability        |  fork() is Standard. fork1()|
   |                             |  and forkall() are Stable.  |
   |_____________________________|_____________________________|
   | MT-Level                    | Async-Signal-Safe.          |
   |_____________________________|_____________________________|

SEE ALSO
     alarm(2),   exec(2),   exit(2),   fcntl(2),    getitimer(2),
     getrlimit(2),   memcntl(2),  mmap(2),  nice(2), priocntl(2),
     semop(2), shmop(2), times(2), umask(2),  door_create(3DOOR),
     exit(3C), plock(3C), pthread_atfork(3C), pthread_create(3C),
     signal(3C),  system(3C),  thr_create(3C)  timer_create(3RT),
     wait(3C),        contract(4),       process(4)attributes(5),
     privileges(5), standards(5)

NOTES
     An applications should call _exit() rather than exit(3C)  if
     it  cannot execve(), since exit() will flush and close stan-
     dard I/O channels and thereby corrupt the  parent  process's
     standard I/O data structures. Using exit(3C) will flush buf-
     fered data twice. See exit(2).

     The thread in the child that calls fork()  or  fork1()  must
     not  depend  on any resources held by threads that no longer
     exist in the child.  In  particular,  locks  held  by  these
     threads will not be released.

     In a multithreaded process,  forkall()  in  one  thread  can
     cause  blocking  system  calls  to be interrupted and return
     with an EINTR error.

No comments:

Post a Comment