Add internal API task_reparent(), used in posix_spawn(). Move libc/spawn/lib_ps.c to sched/task_posixspawn.c; Move libc/spawn/spawn.h to include/nuttx/spawn.h
git-svn-id: svn://svn.code.sf.net/p/nuttx/code/trunk@5531 42af7a65-404d-4744-a932-0658087f49c3
This commit is contained in:
parent
40a93fdac9
commit
6ab00f4a93
@ -124,7 +124,7 @@ static void bultin_semtake(FAR sem_t *sem)
|
||||
do
|
||||
{
|
||||
ret = sem_wait(sem);
|
||||
ASSERT(ret == 0 || errno == EINTR);
|
||||
ASSERT(ret == 0 || get_errno() == EINTR);
|
||||
}
|
||||
while (ret != 0);
|
||||
}
|
||||
@ -152,7 +152,7 @@ static int builtin_taskcreate(int index, FAR const char **argv)
|
||||
|
||||
if (b == NULL)
|
||||
{
|
||||
errno = ENOENT;
|
||||
set_errno(ENOENT);
|
||||
return ERROR;
|
||||
}
|
||||
|
||||
@ -228,7 +228,7 @@ static int builtin_proxy(int argc, char *argv[])
|
||||
{
|
||||
/* Remember the errno value. ret is already set to ERROR */
|
||||
|
||||
g_builtin_parms.errcode = errno;
|
||||
g_builtin_parms.errcode = get_errno();
|
||||
sdbg("ERROR: open of %s failed: %d\n",
|
||||
g_builtin_parms.redirfile, g_builtin_parms.errcode);
|
||||
}
|
||||
@ -246,7 +246,7 @@ static int builtin_proxy(int argc, char *argv[])
|
||||
ret = dup2(fd, 1);
|
||||
if (ret < 0)
|
||||
{
|
||||
g_builtin_parms.errcode = errno;
|
||||
g_builtin_parms.errcode = get_errno();
|
||||
sdbg("ERROR: dup2 failed: %d\n", g_builtin_parms.errcode);
|
||||
}
|
||||
|
||||
@ -266,12 +266,18 @@ static int builtin_proxy(int argc, char *argv[])
|
||||
ret = builtin_taskcreate(g_builtin_parms.index, g_builtin_parms.argv);
|
||||
if (ret < 0)
|
||||
{
|
||||
g_builtin_parms.errcode = errno;
|
||||
g_builtin_parms.errcode = get_errno();
|
||||
sdbg("ERROR: builtin_taskcreate failed: %d\n",
|
||||
g_builtin_parms.errcode);
|
||||
}
|
||||
}
|
||||
|
||||
/* NOTE: There is a logical error here if CONFIG_SCHED_HAVE_PARENT is
|
||||
* defined: The new task is the child of this proxy task, not the
|
||||
* original caller. As a consequence, operations like waitpid() will
|
||||
* fail on the caller's thread.
|
||||
*/
|
||||
|
||||
/* Post the semaphore to inform the parent task that we have completed
|
||||
* what we need to do.
|
||||
*/
|
||||
@ -340,11 +346,21 @@ static inline int builtin_startproxy(int index, FAR const char **argv,
|
||||
ret = sched_getparam(0, ¶m);
|
||||
if (ret < 0)
|
||||
{
|
||||
errcode = errno;
|
||||
errcode = get_errno();
|
||||
sdbg("ERROR: sched_getparam failed: %d\n", errcode);
|
||||
goto errout;
|
||||
goto errout_with_sem;
|
||||
}
|
||||
|
||||
/* Disable pre-emption so that the proxy does not run until we waitpid
|
||||
* is called. This is probably unnecessary since the builtin_proxy has
|
||||
* the same priority as this thread; it should be schedule behind this
|
||||
* task in the ready-to-run list.
|
||||
*/
|
||||
|
||||
#ifdef CONFIG_SCHED_WAITPID
|
||||
sched_lock();
|
||||
#endif
|
||||
|
||||
/* Start the intermediary/proxy task at the same priority as the parent task. */
|
||||
|
||||
proxy = TASK_CREATE("builtin_proxy", param.sched_priority,
|
||||
@ -352,9 +368,9 @@ static inline int builtin_startproxy(int index, FAR const char **argv,
|
||||
(FAR const char **)NULL);
|
||||
if (proxy < 0)
|
||||
{
|
||||
errcode = errno;
|
||||
errcode = get_errno();
|
||||
sdbg("ERROR: Failed to start builtin_proxy: %d\n", errcode);
|
||||
goto errout;
|
||||
goto errout_with_lock;
|
||||
}
|
||||
|
||||
/* Wait for the proxy to complete its job. We could use waitpid()
|
||||
@ -365,7 +381,8 @@ static inline int builtin_startproxy(int index, FAR const char **argv,
|
||||
ret = waitpid(proxy, &status, 0);
|
||||
if (ret < 0)
|
||||
{
|
||||
sdbg("ERROR: waitpid() failed: %d\n", errno);
|
||||
sdbg("ERROR: waitpid() failed: %d\n", get_errno());
|
||||
goto errout_with_lock;
|
||||
}
|
||||
#else
|
||||
bultin_semtake(&g_builtin_execsem);
|
||||
@ -377,7 +394,12 @@ static inline int builtin_startproxy(int index, FAR const char **argv,
|
||||
builtin_semgive(&g_builtin_parmsem);
|
||||
return g_builtin_parms.result;
|
||||
|
||||
errout:
|
||||
errout_with_lock:
|
||||
#ifdef CONFIG_SCHED_WAITPID
|
||||
sched_unlock();
|
||||
#endif
|
||||
|
||||
errout_with_sem:
|
||||
set_errno(errcode);
|
||||
builtin_semgive(&g_builtin_parmsem);
|
||||
return ERROR;
|
||||
|
@ -131,12 +131,14 @@ int nsh_builtin(FAR struct nsh_vtbl_s *vtbl, FAR const char *cmd,
|
||||
{
|
||||
/* The application was successfully started with pre-emption disabled.
|
||||
* In the simplest cases, the application will not have run because the
|
||||
* the scheduler is locked. but in the case were I/O redirected, a
|
||||
* proxy task ran and, as result, so may have the application.
|
||||
* the scheduler is locked. But in the case where I/O was redirected, a
|
||||
* proxy task ran and broke our lock. As result, the application may
|
||||
* have aso ran if its priority was higher than than the priority of
|
||||
* this thread.
|
||||
*
|
||||
* If the application did not run and if the application was not
|
||||
* backgrounded, then we need to wait here for the application to
|
||||
* exit. This only works works with the following options:
|
||||
* If the application did not run to completion and if the application
|
||||
* was not backgrounded, then we need to wait here for the application
|
||||
* to exit. This only works works with the following options:
|
||||
*
|
||||
* - CONFIG_NSH_DISABLEBG - Do not run commands in background
|
||||
* - CONFIG_SCHED_WAITPID - Required to run external commands in
|
||||
@ -156,11 +158,17 @@ int nsh_builtin(FAR struct nsh_vtbl_s *vtbl, FAR const char *cmd,
|
||||
{
|
||||
int rc = 0;
|
||||
|
||||
/* Wait for the application to exit. We did locked the scheduler
|
||||
/* Wait for the application to exit. We did lock the scheduler
|
||||
* above, but that does not guarantee that the application did not
|
||||
* run in the case where I/O was redirected. The scheduler will
|
||||
* be unlocked while waitpid is waiting and if the application has
|
||||
* not yet run, it will be able to to do so.
|
||||
* already run to completion in the case where I/O was redirected.
|
||||
* Here the scheduler will be unlocked while waitpid is waiting
|
||||
* and if the application has not yet run, it will now be able to
|
||||
* do so.
|
||||
*
|
||||
* Also, if CONFIG_SCHED_HAVE_PARENT is defined waitpid() might fail
|
||||
* even if task is still active: If the I/O was re-directed by a
|
||||
* proxy task, then the ask is a child of the proxy, and not this
|
||||
* task. waitpid() fails with ECHILD in either case.
|
||||
*/
|
||||
|
||||
ret = waitpid(ret, &rc, 0);
|
||||
|
@ -168,12 +168,14 @@ int nsh_fileapp(FAR struct nsh_vtbl_s *vtbl, FAR const char *cmd,
|
||||
{
|
||||
/* The application was successfully started with pre-emption disabled.
|
||||
* In the simplest cases, the application will not have run because the
|
||||
* the scheduler is locked. but in the case were I/O redirected, a
|
||||
* proxy task ran and, as result, so may have the application.
|
||||
* the scheduler is locked. But in the case where I/O was redirected, a
|
||||
* proxy task ran and broke our lock. As result, the application may
|
||||
* have aso ran if its priority was higher than than the priority of
|
||||
* this thread.
|
||||
*
|
||||
* If the application did not run and if the application was not
|
||||
* backgrounded, then we need to wait here for the application to
|
||||
* exit. This only works works with the following options:
|
||||
* If the application did not run to completion and if the application
|
||||
* was not backgrounded, then we need to wait here for the application
|
||||
* to exit. This only works works with the following options:
|
||||
*
|
||||
* - CONFIG_NSH_DISABLEBG - Do not run commands in background
|
||||
* - CONFIG_SCHED_WAITPID - Required to run external commands in
|
||||
@ -193,11 +195,12 @@ int nsh_fileapp(FAR struct nsh_vtbl_s *vtbl, FAR const char *cmd,
|
||||
{
|
||||
int rc = 0;
|
||||
|
||||
/* Wait for the application to exit. We did locked the scheduler
|
||||
/* Wait for the application to exit. We did lock the scheduler
|
||||
* above, but that does not guarantee that the application did not
|
||||
* run in the case where I/O was redirected. The scheduler will
|
||||
* be unlocked while waitpid is waiting and if the application has
|
||||
* not yet run, it will be able to to do so.
|
||||
* already run to completion in the case where I/O was redirected.
|
||||
* Here the scheduler will be unlocked while waitpid is waiting
|
||||
* and if the application has not yet run, it will now be able to
|
||||
* do so.
|
||||
*/
|
||||
|
||||
ret = waitpid(pid, &rc, 0);
|
||||
|
Loading…
Reference in New Issue
Block a user