int
sys__lwp_continue(struct lwp *l, const struct sys__lwp_continue_args *uap,
register_t *retval)
{
/* {
syscallarg(lwpid_t) target;
} */
int error;
struct proc *p = l->l_proc;
struct lwp *t;
error = 0;
mutex_enter(p->p_lock);
if ((t = lwp_find(p, SCARG(uap, target))) == NULL) {
mutex_exit(p->p_lock);
return ESRCH;
}
lwp_lock(t);
lwp_continue(t);
mutex_exit(p->p_lock);
return error;
}
void
sigsuspendsetup(struct lwp *l, const sigset_t *ss)
{
struct proc *p = l->l_proc;
/*
* When returning from sigsuspend/pselect/pollts, we want
* the old mask to be restored after the
* signal handler has finished. Thus, we
* save it here and mark the sigctx structure
* to indicate this.
*/
mutex_enter(p->p_lock);
l->l_sigrestore = 1;
l->l_sigoldmask = l->l_sigmask;
l->l_sigmask = *ss;
sigminusset(&sigcantmask, &l->l_sigmask);
/* Check for pending signals when sleeping. */
if (sigispending(l, 0)) {
lwp_lock(l);
l->l_flag |= LW_PENDSIG;
lwp_unlock(l);
}
mutex_exit(p->p_lock);
}
int
sys__lwp_getname(struct lwp *l, const struct sys__lwp_getname_args *uap,
register_t *retval)
{
/* {
syscallarg(lwpid_t) target;
syscallarg(char *) name;
syscallarg(size_t) len;
} */
char name[MAXCOMLEN];
lwpid_t target;
proc_t *p;
lwp_t *t;
if ((target = SCARG(uap, target)) == 0)
target = l->l_lid;
p = curproc;
mutex_enter(p->p_lock);
if ((t = lwp_find(p, target)) == NULL) {
mutex_exit(p->p_lock);
return ESRCH;
}
lwp_lock(t);
if (t->l_name == NULL)
name[0] = '\0';
else
strcpy(name, t->l_name);
lwp_unlock(t);
mutex_exit(p->p_lock);
return copyoutstr(name, SCARG(uap, name), SCARG(uap, len), NULL);
}
int
lwp_unpark(lwpid_t target, const void *hint)
{
sleepq_t *sq;
wchan_t wchan;
kmutex_t *mp;
proc_t *p;
lwp_t *t;
/*
* Easy case: search for the LWP on the sleep queue. If
* it's parked, remove it from the queue and set running.
*/
p = curproc;
wchan = lwp_park_wchan(p, hint);
sq = sleeptab_lookup(&lwp_park_tab, wchan, &mp);
TAILQ_FOREACH(t, sq, l_sleepchain)
if (t->l_proc == p && t->l_lid == target)
break;
if (__predict_true(t != NULL)) {
sleepq_remove(sq, t);
mutex_spin_exit(mp);
return 0;
}
/*
* The LWP hasn't parked yet. Take the hit and mark the
* operation as pending.
*/
mutex_spin_exit(mp);
mutex_enter(p->p_lock);
if ((t = lwp_find(p, target)) == NULL) {
mutex_exit(p->p_lock);
return ESRCH;
}
/*
* It may not have parked yet, we may have raced, or it
* is parked on a different user sync object.
*/
lwp_lock(t);
if (t->l_syncobj == &lwp_park_sobj) {
/* Releases the LWP lock. */
lwp_unsleep(t, true);
} else {
/*
* Set the operation pending. The next call to _lwp_park
* will return early.
*/
t->l_flag |= LW_UNPARKED;
lwp_unlock(t);
}
mutex_exit(p->p_lock);
return 0;
}
int
lwp_park(struct timespec *ts, const void *hint)
{
struct timespec tsx;
sleepq_t *sq;
kmutex_t *mp;
wchan_t wchan;
int timo, error;
lwp_t *l;
/* Fix up the given timeout value. */
if (ts != NULL) {
getnanotime(&tsx);
timespecsub(ts, &tsx, &tsx);
if (tsx.tv_sec < 0 || (tsx.tv_sec == 0 && tsx.tv_nsec <= 0))
return ETIMEDOUT;
if ((error = itimespecfix(&tsx)) != 0)
return error;
timo = tstohz(&tsx);
KASSERT(timo != 0);
} else
timo = 0;
/* Find and lock the sleep queue. */
l = curlwp;
wchan = lwp_park_wchan(l->l_proc, hint);
sq = sleeptab_lookup(&lwp_park_tab, wchan, &mp);
/*
* Before going the full route and blocking, check to see if an
* unpark op is pending.
*/
lwp_lock(l);
if ((l->l_flag & (LW_CANCELLED | LW_UNPARKED)) != 0) {
l->l_flag &= ~(LW_CANCELLED | LW_UNPARKED);
lwp_unlock(l);
mutex_spin_exit(mp);
return EALREADY;
}
lwp_unlock_to(l, mp);
l->l_biglocks = 0;
sleepq_enqueue(sq, wchan, "parked", &lwp_park_sobj);
error = sleepq_block(timo, true);
switch (error) {
case EWOULDBLOCK:
error = ETIMEDOUT;
break;
case ERESTART:
error = EINTR;
break;
default:
/* nothing */
break;
}
return error;
}
int
do_lwp_create(lwp_t *l, void *arg, u_long flags, lwpid_t *new_lwp)
{
struct proc *p = l->l_proc;
struct lwp *l2;
struct schedstate_percpu *spc;
vaddr_t uaddr;
int error;
/* XXX check against resource limits */
uaddr = uvm_uarea_alloc();
if (__predict_false(uaddr == 0))
return ENOMEM;
error = lwp_create(l, p, uaddr, flags & LWP_DETACHED,
NULL, 0, p->p_emul->e_startlwp, arg, &l2, l->l_class);
if (__predict_false(error)) {
uvm_uarea_free(uaddr);
return error;
}
*new_lwp = l2->l_lid;
/*
* Set the new LWP running, unless the caller has requested that
* it be created in suspended state. If the process is stopping,
* then the LWP is created stopped.
*/
mutex_enter(p->p_lock);
lwp_lock(l2);
spc = &l2->l_cpu->ci_schedstate;
if ((flags & LWP_SUSPENDED) == 0 &&
(l->l_flag & (LW_WREBOOT | LW_WSUSPEND | LW_WEXIT)) == 0) {
if (p->p_stat == SSTOP || (p->p_sflag & PS_STOPPING) != 0) {
KASSERT(l2->l_wchan == NULL);
l2->l_stat = LSSTOP;
p->p_nrlwps--;
lwp_unlock_to(l2, spc->spc_lwplock);
} else {
KASSERT(lwp_locked(l2, spc->spc_mutex));
l2->l_stat = LSRUN;
sched_enqueue(l2, false);
lwp_unlock(l2);
}
} else {
l2->l_stat = LSSUSPENDED;
p->p_nrlwps--;
lwp_unlock_to(l2, spc->spc_lwplock);
}
mutex_exit(p->p_lock);
return 0;
}
int
lwp_park(clockid_t clock_id, int flags, struct timespec *ts, const void *hint)
{
sleepq_t *sq;
kmutex_t *mp;
wchan_t wchan;
int timo, error;
lwp_t *l;
if (ts != NULL) {
if ((error = ts2timo(clock_id, flags, ts, &timo, NULL)) != 0)
return error;
KASSERT(timo != 0);
} else {
timo = 0;
}
/* Find and lock the sleep queue. */
l = curlwp;
wchan = lwp_park_wchan(l->l_proc, hint);
sq = sleeptab_lookup(&lwp_park_tab, wchan, &mp);
/*
* Before going the full route and blocking, check to see if an
* unpark op is pending.
*/
lwp_lock(l);
if ((l->l_flag & (LW_CANCELLED | LW_UNPARKED)) != 0) {
l->l_flag &= ~(LW_CANCELLED | LW_UNPARKED);
lwp_unlock(l);
mutex_spin_exit(mp);
return EALREADY;
}
lwp_unlock_to(l, mp);
l->l_biglocks = 0;
sleepq_enqueue(sq, wchan, "parked", &lwp_park_sobj);
error = sleepq_block(timo, true);
switch (error) {
case EWOULDBLOCK:
error = ETIMEDOUT;
break;
case ERESTART:
error = EINTR;
break;
default:
/* nothing */
break;
}
return error;
}
int
sys__lwp_setname(struct lwp *l, const struct sys__lwp_setname_args *uap,
register_t *retval)
{
/* {
syscallarg(lwpid_t) target;
syscallarg(const char *) name;
} */
char *name, *oname;
lwpid_t target;
proc_t *p;
lwp_t *t;
int error;
if ((target = SCARG(uap, target)) == 0)
target = l->l_lid;
name = kmem_alloc(MAXCOMLEN, KM_SLEEP);
if (name == NULL)
return ENOMEM;
error = copyinstr(SCARG(uap, name), name, MAXCOMLEN, NULL);
switch (error) {
case ENAMETOOLONG:
case 0:
name[MAXCOMLEN - 1] = '\0';
break;
default:
kmem_free(name, MAXCOMLEN);
return error;
}
p = curproc;
mutex_enter(p->p_lock);
if ((t = lwp_find(p, target)) == NULL) {
mutex_exit(p->p_lock);
kmem_free(name, MAXCOMLEN);
return ESRCH;
}
lwp_lock(t);
oname = t->l_name;
t->l_name = name;
lwp_unlock(t);
mutex_exit(p->p_lock);
if (oname != NULL)
kmem_free(oname, MAXCOMLEN);
return 0;
}
void
sigsuspendteardown(struct lwp *l)
{
struct proc *p = l->l_proc;
mutex_enter(p->p_lock);
/* Check for pending signals when sleeping. */
if (l->l_sigrestore) {
if (sigispending(l, 0)) {
lwp_lock(l);
l->l_flag |= LW_PENDSIG;
lwp_unlock(l);
} else {
l->l_sigrestore = 0;
l->l_sigmask = l->l_sigoldmask;
}
}
mutex_exit(p->p_lock);
}
int
sigprocmask1(struct lwp *l, int how, const sigset_t *nss, sigset_t *oss)
{
sigset_t *mask = &l->l_sigmask;
bool more;
KASSERT(mutex_owned(l->l_proc->p_lock));
if (oss) {
*oss = *mask;
}
if (nss == NULL) {
return 0;
}
switch (how) {
case SIG_BLOCK:
sigplusset(nss, mask);
more = false;
break;
case SIG_UNBLOCK:
sigminusset(nss, mask);
more = true;
break;
case SIG_SETMASK:
*mask = *nss;
more = true;
break;
default:
return EINVAL;
}
sigminusset(&sigcantmask, mask);
if (more && sigispending(l, 0)) {
/*
* Check for pending signals on return to user.
*/
lwp_lock(l);
l->l_flag |= LW_PENDSIG;
lwp_unlock(l);
}
return 0;
}
/*
* sleepq_timeout:
*
* Entered via the callout(9) subsystem to time out an LWP that is on a
* sleep queue.
*/
void
sleepq_timeout(void *arg)
{
lwp_t *l = arg;
/*
* Lock the LWP. Assuming it's still on the sleep queue, its
* current mutex will also be the sleep queue mutex.
*/
lwp_lock(l);
if (l->l_wchan == NULL) {
/* Somebody beat us to it. */
lwp_unlock(l);
return;
}
//lwp_unsleep(l, true);
sleepq_unsleep(l, true);
}
int
sys__lwp_wakeup(struct lwp *l, const struct sys__lwp_wakeup_args *uap,
register_t *retval)
{
/* {
syscallarg(lwpid_t) target;
} */
struct lwp *t;
struct proc *p;
int error;
p = l->l_proc;
mutex_enter(p->p_lock);
if ((t = lwp_find(p, SCARG(uap, target))) == NULL) {
mutex_exit(p->p_lock);
return ESRCH;
}
lwp_lock(t);
t->l_flag |= (LW_CANCELLED | LW_UNPARKED);
if (t->l_stat != LSSLEEP) {
lwp_unlock(t);
error = ENODEV;
} else if ((t->l_flag & LW_SINTR) == 0) {
lwp_unlock(t);
error = EBUSY;
} else {
/* Wake it up. lwp_unsleep() will release the LWP lock. */
lwp_unsleep(t, true);
error = 0;
}
mutex_exit(p->p_lock);
return error;
}
int
sys__lwp_unpark_all(struct lwp *l, const struct sys__lwp_unpark_all_args *uap,
register_t *retval)
{
/* {
syscallarg(const lwpid_t *) targets;
syscallarg(size_t) ntargets;
syscallarg(const void *) hint;
} */
struct proc *p;
struct lwp *t;
sleepq_t *sq;
wchan_t wchan;
lwpid_t targets[32], *tp, *tpp, *tmax, target;
int error;
kmutex_t *mp;
u_int ntargets;
size_t sz;
p = l->l_proc;
ntargets = SCARG(uap, ntargets);
if (SCARG(uap, targets) == NULL) {
/*
* Let the caller know how much we are willing to do, and
* let it unpark the LWPs in blocks.
*/
*retval = LWP_UNPARK_MAX;
return 0;
}
if (ntargets > LWP_UNPARK_MAX || ntargets == 0)
return EINVAL;
/*
* Copy in the target array. If it's a small number of LWPs, then
* place the numbers on the stack.
*/
sz = sizeof(target) * ntargets;
if (sz <= sizeof(targets))
tp = targets;
else {
tp = kmem_alloc(sz, KM_SLEEP);
if (tp == NULL)
return ENOMEM;
}
error = copyin(SCARG(uap, targets), tp, sz);
if (error != 0) {
if (tp != targets) {
kmem_free(tp, sz);
}
return error;
}
wchan = lwp_park_wchan(p, SCARG(uap, hint));
sq = sleeptab_lookup(&lwp_park_tab, wchan, &mp);
for (tmax = tp + ntargets, tpp = tp; tpp < tmax; tpp++) {
target = *tpp;
/*
* Easy case: search for the LWP on the sleep queue. If
* it's parked, remove it from the queue and set running.
*/
TAILQ_FOREACH(t, sq, l_sleepchain)
if (t->l_proc == p && t->l_lid == target)
break;
if (t != NULL) {
sleepq_remove(sq, t);
continue;
}
/*
* The LWP hasn't parked yet. Take the hit and
* mark the operation as pending.
*/
mutex_spin_exit(mp);
mutex_enter(p->p_lock);
if ((t = lwp_find(p, target)) == NULL) {
mutex_exit(p->p_lock);
mutex_spin_enter(mp);
continue;
}
lwp_lock(t);
/*
* It may not have parked yet, we may have raced, or
* it is parked on a different user sync object.
*/
if (t->l_syncobj == &lwp_park_sobj) {
/* Releases the LWP lock. */
lwp_unsleep(t, true);
} else {
/*
* Set the operation pending. The next call to
* _lwp_park will return early.
*/
t->l_flag |= LW_UNPARKED;
lwp_unlock(t);
}
mutex_exit(p->p_lock);
mutex_spin_enter(mp);
}
mutex_spin_exit(mp);
if (tp != targets)
kmem_free(tp, sz);
return 0;
}
/*
* General fork call. Note that another LWP in the process may call exec()
* or exit() while we are forking. It's safe to continue here, because
* neither operation will complete until all LWPs have exited the process.
*/
int
fork1(struct lwp *l1, int flags, int exitsig, void *stack, size_t stacksize,
void (*func)(void *), void *arg, register_t *retval,
struct proc **rnewprocp)
{
struct proc *p1, *p2, *parent;
struct plimit *p1_lim;
uid_t uid;
struct lwp *l2;
int count;
vaddr_t uaddr;
int tnprocs;
int tracefork;
int error = 0;
p1 = l1->l_proc;
uid = kauth_cred_getuid(l1->l_cred);
tnprocs = atomic_inc_uint_nv(&nprocs);
/*
* Although process entries are dynamically created, we still keep
* a global limit on the maximum number we will create.
*/
if (__predict_false(tnprocs >= maxproc))
error = -1;
else
error = kauth_authorize_process(l1->l_cred,
KAUTH_PROCESS_FORK, p1, KAUTH_ARG(tnprocs), NULL, NULL);
if (error) {
static struct timeval lasttfm;
atomic_dec_uint(&nprocs);
if (ratecheck(&lasttfm, &fork_tfmrate))
tablefull("proc", "increase kern.maxproc or NPROC");
if (forkfsleep)
kpause("forkmx", false, forkfsleep, NULL);
return EAGAIN;
}
/*
* Enforce limits.
*/
count = chgproccnt(uid, 1);
if (__predict_false(count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
if (kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_RLIMIT,
p1, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
&p1->p_rlimit[RLIMIT_NPROC], KAUTH_ARG(RLIMIT_NPROC)) != 0) {
(void)chgproccnt(uid, -1);
atomic_dec_uint(&nprocs);
if (forkfsleep)
kpause("forkulim", false, forkfsleep, NULL);
return EAGAIN;
}
}
/*
* Allocate virtual address space for the U-area now, while it
* is still easy to abort the fork operation if we're out of
* kernel virtual address space.
*/
uaddr = uvm_uarea_alloc();
if (__predict_false(uaddr == 0)) {
(void)chgproccnt(uid, -1);
atomic_dec_uint(&nprocs);
return ENOMEM;
}
/*
* We are now committed to the fork. From here on, we may
* block on resources, but resource allocation may NOT fail.
*/
/* Allocate new proc. */
p2 = proc_alloc();
/*
* Make a proc table entry for the new process.
* Start by zeroing the section of proc that is zero-initialized,
* then copy the section that is copied directly from the parent.
*/
memset(&p2->p_startzero, 0,
(unsigned) ((char *)&p2->p_endzero - (char *)&p2->p_startzero));
memcpy(&p2->p_startcopy, &p1->p_startcopy,
(unsigned) ((char *)&p2->p_endcopy - (char *)&p2->p_startcopy));
TAILQ_INIT(&p2->p_sigpend.sp_info);
LIST_INIT(&p2->p_lwps);
LIST_INIT(&p2->p_sigwaiters);
/*
* Duplicate sub-structures as needed.
* Increase reference counts on shared objects.
* Inherit flags we want to keep. The flags related to SIGCHLD
* handling are important in order to keep a consistent behaviour
* for the child after the fork. If we are a 32-bit process, the
* child will be too.
*/
p2->p_flag =
p1->p_flag & (PK_SUGID | PK_NOCLDWAIT | PK_CLDSIGIGN | PK_32);
p2->p_emul = p1->p_emul;
p2->p_execsw = p1->p_execsw;
if (flags & FORK_SYSTEM) {
/*
* Mark it as a system process. Set P_NOCLDWAIT so that
* children are reparented to init(8) when they exit.
* init(8) can easily wait them out for us.
*/
p2->p_flag |= (PK_SYSTEM | PK_NOCLDWAIT);
}
mutex_init(&p2->p_stmutex, MUTEX_DEFAULT, IPL_HIGH);
mutex_init(&p2->p_auxlock, MUTEX_DEFAULT, IPL_NONE);
rw_init(&p2->p_reflock);
cv_init(&p2->p_waitcv, "wait");
cv_init(&p2->p_lwpcv, "lwpwait");
/*
* Share a lock between the processes if they are to share signal
* state: we must synchronize access to it.
*/
if (flags & FORK_SHARESIGS) {
p2->p_lock = p1->p_lock;
mutex_obj_hold(p1->p_lock);
} else
p2->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
kauth_proc_fork(p1, p2);
p2->p_raslist = NULL;
#if defined(__HAVE_RAS)
ras_fork(p1, p2);
#endif
/* bump references to the text vnode (for procfs) */
p2->p_textvp = p1->p_textvp;
if (p2->p_textvp)
vref(p2->p_textvp);
if (flags & FORK_SHAREFILES)
fd_share(p2);
else if (flags & FORK_CLEANFILES)
p2->p_fd = fd_init(NULL);
else
p2->p_fd = fd_copy();
/* XXX racy */
p2->p_mqueue_cnt = p1->p_mqueue_cnt;
if (flags & FORK_SHARECWD)
cwdshare(p2);
else
p2->p_cwdi = cwdinit();
/*
* Note: p_limit (rlimit stuff) is copy-on-write, so normally
* we just need increase pl_refcnt.
*/
p1_lim = p1->p_limit;
if (!p1_lim->pl_writeable) {
lim_addref(p1_lim);
p2->p_limit = p1_lim;
} else {
p2->p_limit = lim_copy(p1_lim);
}
if (flags & FORK_PPWAIT) {
/* Mark ourselves as waiting for a child. */
l1->l_pflag |= LP_VFORKWAIT;
p2->p_lflag = PL_PPWAIT;
p2->p_vforklwp = l1;
} else {
p2->p_lflag = 0;
}
p2->p_sflag = 0;
p2->p_slflag = 0;
parent = (flags & FORK_NOWAIT) ? initproc : p1;
p2->p_pptr = parent;
p2->p_ppid = parent->p_pid;
LIST_INIT(&p2->p_children);
p2->p_aio = NULL;
#ifdef KTRACE
/*
* Copy traceflag and tracefile if enabled.
* If not inherited, these were zeroed above.
*/
if (p1->p_traceflag & KTRFAC_INHERIT) {
mutex_enter(&ktrace_lock);
p2->p_traceflag = p1->p_traceflag;
if ((p2->p_tracep = p1->p_tracep) != NULL)
ktradref(p2);
mutex_exit(&ktrace_lock);
}
#endif
/*
* Create signal actions for the child process.
*/
p2->p_sigacts = sigactsinit(p1, flags & FORK_SHARESIGS);
mutex_enter(p1->p_lock);
p2->p_sflag |=
(p1->p_sflag & (PS_STOPFORK | PS_STOPEXEC | PS_NOCLDSTOP));
sched_proc_fork(p1, p2);
mutex_exit(p1->p_lock);
p2->p_stflag = p1->p_stflag;
/*
* p_stats.
* Copy parts of p_stats, and zero out the rest.
*/
p2->p_stats = pstatscopy(p1->p_stats);
/*
* Set up the new process address space.
*/
uvm_proc_fork(p1, p2, (flags & FORK_SHAREVM) ? true : false);
/*
* Finish creating the child process.
* It will return through a different path later.
*/
lwp_create(l1, p2, uaddr, (flags & FORK_PPWAIT) ? LWP_VFORK : 0,
stack, stacksize, (func != NULL) ? func : child_return, arg, &l2,
l1->l_class);
/*
* Inherit l_private from the parent.
* Note that we cannot use lwp_setprivate() here since that
* also sets the CPU TLS register, which is incorrect if the
* process has changed that without letting the kernel know.
*/
l2->l_private = l1->l_private;
/*
* If emulation has a process fork hook, call it now.
*/
if (p2->p_emul->e_proc_fork)
(*p2->p_emul->e_proc_fork)(p2, l1, flags);
/*
* ...and finally, any other random fork hooks that subsystems
* might have registered.
*/
doforkhooks(p2, p1);
SDT_PROBE(proc,,,create, p2, p1, flags, 0, 0);
/*
* It's now safe for the scheduler and other processes to see the
* child process.
*/
mutex_enter(proc_lock);
if (p1->p_session->s_ttyvp != NULL && p1->p_lflag & PL_CONTROLT)
p2->p_lflag |= PL_CONTROLT;
LIST_INSERT_HEAD(&parent->p_children, p2, p_sibling);
p2->p_exitsig = exitsig; /* signal for parent on exit */
/*
* We don't want to tracefork vfork()ed processes because they
* will not receive the SIGTRAP until it is too late.
*/
tracefork = (p1->p_slflag & (PSL_TRACEFORK|PSL_TRACED)) ==
(PSL_TRACEFORK|PSL_TRACED) && (flags && FORK_PPWAIT) == 0;
if (tracefork) {
p2->p_slflag |= PSL_TRACED;
p2->p_opptr = p2->p_pptr;
if (p2->p_pptr != p1->p_pptr) {
struct proc *parent1 = p2->p_pptr;
if (parent1->p_lock < p2->p_lock) {
if (!mutex_tryenter(parent1->p_lock)) {
mutex_exit(p2->p_lock);
mutex_enter(parent1->p_lock);
}
} else if (parent1->p_lock > p2->p_lock) {
mutex_enter(parent1->p_lock);
}
parent1->p_slflag |= PSL_CHTRACED;
proc_reparent(p2, p1->p_pptr);
if (parent1->p_lock != p2->p_lock)
mutex_exit(parent1->p_lock);
}
/*
* Set ptrace status.
*/
p1->p_fpid = p2->p_pid;
p2->p_fpid = p1->p_pid;
}
LIST_INSERT_AFTER(p1, p2, p_pglist);
LIST_INSERT_HEAD(&allproc, p2, p_list);
p2->p_trace_enabled = trace_is_enabled(p2);
#ifdef __HAVE_SYSCALL_INTERN
(*p2->p_emul->e_syscall_intern)(p2);
#endif
/*
* Update stats now that we know the fork was successful.
*/
uvmexp.forks++;
if (flags & FORK_PPWAIT)
uvmexp.forks_ppwait++;
if (flags & FORK_SHAREVM)
uvmexp.forks_sharevm++;
/*
* Pass a pointer to the new process to the caller.
*/
if (rnewprocp != NULL)
*rnewprocp = p2;
if (ktrpoint(KTR_EMUL))
p2->p_traceflag |= KTRFAC_TRC_EMUL;
/*
* Notify any interested parties about the new process.
*/
if (!SLIST_EMPTY(&p1->p_klist)) {
mutex_exit(proc_lock);
KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
mutex_enter(proc_lock);
}
/*
* Make child runnable, set start time, and add to run queue except
* if the parent requested the child to start in SSTOP state.
*/
mutex_enter(p2->p_lock);
/*
* Start profiling.
*/
if ((p2->p_stflag & PST_PROFIL) != 0) {
mutex_spin_enter(&p2->p_stmutex);
startprofclock(p2);
mutex_spin_exit(&p2->p_stmutex);
}
getmicrotime(&p2->p_stats->p_start);
p2->p_acflag = AFORK;
lwp_lock(l2);
KASSERT(p2->p_nrlwps == 1);
if (p2->p_sflag & PS_STOPFORK) {
struct schedstate_percpu *spc = &l2->l_cpu->ci_schedstate;
p2->p_nrlwps = 0;
p2->p_stat = SSTOP;
p2->p_waited = 0;
p1->p_nstopchild++;
l2->l_stat = LSSTOP;
KASSERT(l2->l_wchan == NULL);
lwp_unlock_to(l2, spc->spc_lwplock);
} else {
p2->p_nrlwps = 1;
p2->p_stat = SACTIVE;
l2->l_stat = LSRUN;
sched_enqueue(l2, false);
lwp_unlock(l2);
}
/*
* Return child pid to parent process,
* marking us as parent via retval[1].
*/
if (retval != NULL) {
retval[0] = p2->p_pid;
retval[1] = 0;
}
mutex_exit(p2->p_lock);
/*
* Preserve synchronization semantics of vfork. If waiting for
* child to exec or exit, sleep until it clears LP_VFORKWAIT.
*/
#if 0
while (l1->l_pflag & LP_VFORKWAIT) {
cv_wait(&l1->l_waitcv, proc_lock);
}
#else
while (p2->p_lflag & PL_PPWAIT)
cv_wait(&p1->p_waitcv, proc_lock);
#endif
/*
* Let the parent know that we are tracing its child.
*/
if (tracefork) {
ksiginfo_t ksi;
KSI_INIT_EMPTY(&ksi);
ksi.ksi_signo = SIGTRAP;
ksi.ksi_lid = l1->l_lid;
kpsignal(p1, &ksi, NULL);
}
mutex_exit(proc_lock);
return 0;
}
/*
* Fork a kernel thread. Any process can request this to be done.
*/
int
kthread_create(pri_t pri, int flag, struct cpu_info *ci,
void (*func)(void *), void *arg,
lwp_t **lp, const char *fmt, ...)
{
lwp_t *l;
vaddr_t uaddr;
bool inmem;
int error;
va_list ap;
int lc;
inmem = uvm_uarea_alloc(&uaddr);
if (uaddr == 0)
return ENOMEM;
if ((flag & KTHREAD_TS) != 0) {
lc = SCHED_OTHER;
} else {
lc = SCHED_RR;
}
error = lwp_create(&lwp0, &proc0, uaddr, inmem, LWP_DETACHED, NULL,
0, func, arg, &l, lc);
if (error) {
uvm_uarea_free(uaddr, curcpu());
return error;
}
uvm_lwp_hold(l);
if (fmt != NULL) {
l->l_name = kmem_alloc(MAXCOMLEN, KM_SLEEP);
if (l->l_name == NULL) {
lwp_exit(l);
return ENOMEM;
}
va_start(ap, fmt);
vsnprintf(l->l_name, MAXCOMLEN, fmt, ap);
va_end(ap);
}
/*
* Set parameters.
*/
if ((flag & KTHREAD_INTR) != 0) {
KASSERT((flag & KTHREAD_MPSAFE) != 0);
}
if (pri == PRI_NONE) {
if ((flag & KTHREAD_TS) != 0) {
/* Maximum user priority level. */
pri = MAXPRI_USER;
} else {
/* Minimum kernel priority level. */
pri = PRI_KTHREAD;
}
}
mutex_enter(proc0.p_lock);
lwp_lock(l);
l->l_priority = pri;
if (ci != NULL) {
if (ci != l->l_cpu) {
lwp_unlock_to(l, ci->ci_schedstate.spc_mutex);
lwp_lock(l);
}
l->l_pflag |= LP_BOUND;
l->l_cpu = ci;
}
if ((flag & KTHREAD_INTR) != 0)
l->l_pflag |= LP_INTR;
if ((flag & KTHREAD_MPSAFE) == 0)
l->l_pflag &= ~LP_MPSAFE;
/*
* Set the new LWP running, unless the caller has requested
* otherwise.
*/
if ((flag & KTHREAD_IDLE) == 0) {
l->l_stat = LSRUN;
sched_enqueue(l, false);
lwp_unlock(l);
} else
lwp_unlock_to(l, ci->ci_schedstate.spc_lwplock);
/*
* The LWP is not created suspended or stopped and cannot be set
* into those states later, so must be considered runnable.
*/
proc0.p_nrlwps++;
mutex_exit(proc0.p_lock);
/* All done! */
if (lp != NULL)
*lp = l;
return (0);
}
int
sys__lwp_suspend(struct lwp *l, const struct sys__lwp_suspend_args *uap,
register_t *retval)
{
/* {
syscallarg(lwpid_t) target;
} */
struct proc *p = l->l_proc;
struct lwp *t;
int error;
mutex_enter(p->p_lock);
if ((t = lwp_find(p, SCARG(uap, target))) == NULL) {
mutex_exit(p->p_lock);
return ESRCH;
}
/*
* Check for deadlock, which is only possible when we're suspending
* ourself. XXX There is a short race here, as p_nrlwps is only
* incremented when an LWP suspends itself on the kernel/user
* boundary. It's still possible to kill -9 the process so we
* don't bother checking further.
*/
lwp_lock(t);
if ((t == l && p->p_nrlwps == 1) ||
(l->l_flag & (LW_WCORE | LW_WEXIT)) != 0) {
lwp_unlock(t);
mutex_exit(p->p_lock);
return EDEADLK;
}
/*
* Suspend the LWP. XXX If it's on a different CPU, we should wait
* for it to be preempted, where it will put itself to sleep.
*
* Suspension of the current LWP will happen on return to userspace.
*/
error = lwp_suspend(l, t);
if (error) {
mutex_exit(p->p_lock);
return error;
}
/*
* Wait for:
* o process exiting
* o target LWP suspended
* o target LWP not suspended and L_WSUSPEND clear
* o target LWP exited
*/
for (;;) {
error = cv_wait_sig(&p->p_lwpcv, p->p_lock);
if (error) {
error = ERESTART;
break;
}
if (lwp_find(p, SCARG(uap, target)) == NULL) {
error = ESRCH;
break;
}
if ((l->l_flag | t->l_flag) & (LW_WCORE | LW_WEXIT)) {
error = ERESTART;
break;
}
if (t->l_stat == LSSUSPENDED ||
(t->l_flag & LW_WSUSPEND) == 0)
break;
}
mutex_exit(p->p_lock);
return error;
}
int
sigaction1(struct lwp *l, int signum, const struct sigaction *nsa,
struct sigaction *osa, const void *tramp, int vers)
{
struct proc *p;
struct sigacts *ps;
sigset_t tset;
int prop, error;
ksiginfoq_t kq;
static bool v0v1valid;
if (signum <= 0 || signum >= NSIG)
return EINVAL;
p = l->l_proc;
error = 0;
ksiginfo_queue_init(&kq);
/*
* Trampoline ABI version 0 is reserved for the legacy kernel
* provided on-stack trampoline. Conversely, if we are using a
* non-0 ABI version, we must have a trampoline. Only validate the
* vers if a new sigaction was supplied and there was an actual
* handler specified (not SIG_IGN or SIG_DFL), which don't require
* a trampoline. Emulations use legacy kernel trampolines with
* version 0, alternatively check for that too.
*
* If version < 2, we try to autoload the compat module. Note
* that we interlock with the unload check in compat_modcmd()
* using kernconfig_lock. If the autoload fails, we don't try it
* again for this process.
*/
if (nsa != NULL && nsa->sa_handler != SIG_IGN
&& nsa->sa_handler != SIG_DFL) {
if (__predict_false(vers < 2)) {
if (p->p_flag & PK_32)
v0v1valid = true;
else if ((p->p_lflag & PL_SIGCOMPAT) == 0) {
kernconfig_lock();
if (sendsig_sigcontext_vec == NULL) {
(void)module_autoload("compat",
MODULE_CLASS_ANY);
}
if (sendsig_sigcontext_vec != NULL) {
/*
* We need to remember if the
* sigcontext method may be useable,
* because libc may use it even
* if siginfo is available.
*/
v0v1valid = true;
}
mutex_enter(proc_lock);
/*
* Prevent unload of compat module while
* this process remains.
*/
p->p_lflag |= PL_SIGCOMPAT;
mutex_exit(proc_lock);
kernconfig_unlock();
}
}
switch (vers) {
case 0:
/* sigcontext, kernel supplied trampoline. */
if (tramp != NULL || !v0v1valid) {
return EINVAL;
}
break;
case 1:
/* sigcontext, user supplied trampoline. */
if (tramp == NULL || !v0v1valid) {
return EINVAL;
}
break;
case 2:
case 3:
/* siginfo, user supplied trampoline. */
if (tramp == NULL) {
return EINVAL;
}
break;
default:
return EINVAL;
}
}
mutex_enter(p->p_lock);
ps = p->p_sigacts;
if (osa)
*osa = SIGACTION_PS(ps, signum);
if (!nsa)
goto out;
prop = sigprop[signum];
if ((nsa->sa_flags & ~SA_ALLBITS) || (prop & SA_CANTMASK)) {
error = EINVAL;
goto out;
}
SIGACTION_PS(ps, signum) = *nsa;
ps->sa_sigdesc[signum].sd_tramp = tramp;
ps->sa_sigdesc[signum].sd_vers = vers;
sigminusset(&sigcantmask, &SIGACTION_PS(ps, signum).sa_mask);
if ((prop & SA_NORESET) != 0)
SIGACTION_PS(ps, signum).sa_flags &= ~SA_RESETHAND;
if (signum == SIGCHLD) {
if (nsa->sa_flags & SA_NOCLDSTOP)
p->p_sflag |= PS_NOCLDSTOP;
else
p->p_sflag &= ~PS_NOCLDSTOP;
if (nsa->sa_flags & SA_NOCLDWAIT) {
/*
* Paranoia: since SA_NOCLDWAIT is implemented by
* reparenting the dying child to PID 1 (and trust
* it to reap the zombie), PID 1 itself is forbidden
* to set SA_NOCLDWAIT.
*/
if (p->p_pid == 1)
p->p_flag &= ~PK_NOCLDWAIT;
else
p->p_flag |= PK_NOCLDWAIT;
} else
p->p_flag &= ~PK_NOCLDWAIT;
if (nsa->sa_handler == SIG_IGN) {
/*
* Paranoia: same as above.
*/
if (p->p_pid == 1)
p->p_flag &= ~PK_CLDSIGIGN;
else
p->p_flag |= PK_CLDSIGIGN;
} else
p->p_flag &= ~PK_CLDSIGIGN;
}
if ((nsa->sa_flags & SA_NODEFER) == 0)
sigaddset(&SIGACTION_PS(ps, signum).sa_mask, signum);
else
sigdelset(&SIGACTION_PS(ps, signum).sa_mask, signum);
/*
* Set bit in p_sigctx.ps_sigignore for signals that are set to
* SIG_IGN, and for signals set to SIG_DFL where the default is to
* ignore. However, don't put SIGCONT in p_sigctx.ps_sigignore, as
* we have to restart the process.
*/
if (nsa->sa_handler == SIG_IGN ||
(nsa->sa_handler == SIG_DFL && (prop & SA_IGNORE) != 0)) {
/* Never to be seen again. */
sigemptyset(&tset);
sigaddset(&tset, signum);
sigclearall(p, &tset, &kq);
if (signum != SIGCONT) {
/* Easier in psignal */
sigaddset(&p->p_sigctx.ps_sigignore, signum);
}
sigdelset(&p->p_sigctx.ps_sigcatch, signum);
} else {
sigdelset(&p->p_sigctx.ps_sigignore, signum);
if (nsa->sa_handler == SIG_DFL)
sigdelset(&p->p_sigctx.ps_sigcatch, signum);
else
sigaddset(&p->p_sigctx.ps_sigcatch, signum);
}
/*
* Previously held signals may now have become visible. Ensure that
* we check for them before returning to userspace.
*/
if (sigispending(l, 0)) {
lwp_lock(l);
l->l_flag |= LW_PENDSIG;
lwp_unlock(l);
}
out:
mutex_exit(p->p_lock);
ksiginfo_queue_drain(&kq);
return error;
}
/* ARGSUSED */
int
sys__lwp_create(struct lwp *l, const struct sys__lwp_create_args *uap, register_t *retval)
{
/* {
syscallarg(const ucontext_t *) ucp;
syscallarg(u_long) flags;
syscallarg(lwpid_t *) new_lwp;
} */
struct proc *p = l->l_proc;
struct lwp *l2;
vaddr_t uaddr;
bool inmem;
ucontext_t *newuc;
int error, lid;
#ifdef KERN_SA
mutex_enter(p->p_lock);
if ((p->p_sflag & (PS_SA | PS_WEXIT)) != 0 || p->p_sa != NULL) {
mutex_exit(p->p_lock);
return EINVAL;
}
mutex_exit(p->p_lock);
#endif
newuc = pool_get(&lwp_uc_pool, PR_WAITOK);
error = copyin(SCARG(uap, ucp), newuc, p->p_emul->e_ucsize);
if (error) {
pool_put(&lwp_uc_pool, newuc);
return error;
}
/* XXX check against resource limits */
inmem = uvm_uarea_alloc(&uaddr);
if (__predict_false(uaddr == 0)) {
pool_put(&lwp_uc_pool, newuc);
return ENOMEM;
}
error = lwp_create(l, p, uaddr, inmem, SCARG(uap, flags) & LWP_DETACHED,
NULL, 0, p->p_emul->e_startlwp, newuc, &l2, l->l_class);
if (error) {
uvm_uarea_free(uaddr, curcpu());
pool_put(&lwp_uc_pool, newuc);
return error;
}
lid = l2->l_lid;
error = copyout(&lid, SCARG(uap, new_lwp), sizeof(lid));
if (error) {
lwp_exit(l2);
pool_put(&lwp_uc_pool, newuc);
return error;
}
/*
* Set the new LWP running, unless the caller has requested that
* it be created in suspended state. If the process is stopping,
* then the LWP is created stopped.
*/
mutex_enter(p->p_lock);
lwp_lock(l2);
if ((SCARG(uap, flags) & LWP_SUSPENDED) == 0 &&
(l->l_flag & (LW_WREBOOT | LW_WSUSPEND | LW_WEXIT)) == 0) {
if (p->p_stat == SSTOP || (p->p_sflag & PS_STOPPING) != 0)
l2->l_stat = LSSTOP;
else {
KASSERT(lwp_locked(l2, l2->l_cpu->ci_schedstate.spc_mutex));
p->p_nrlwps++;
l2->l_stat = LSRUN;
sched_enqueue(l2, false);
}
lwp_unlock(l2);
} else {
l2->l_stat = LSSUSPENDED;
lwp_unlock_to(l2, l2->l_cpu->ci_schedstate.spc_lwplock);
}
mutex_exit(p->p_lock);
return 0;
}
static int
linux_clone_nptl(struct lwp *l, const struct linux_sys_clone_args *uap, register_t *retval)
{
/* {
syscallarg(int) flags;
syscallarg(void *) stack;
syscallarg(void *) parent_tidptr;
syscallarg(void *) tls;
syscallarg(void *) child_tidptr;
} */
struct proc *p;
struct lwp *l2;
struct linux_emuldata *led;
void *parent_tidptr, *tls, *child_tidptr;
struct schedstate_percpu *spc;
vaddr_t uaddr;
lwpid_t lid;
int flags, tnprocs, error;
p = l->l_proc;
flags = SCARG(uap, flags);
parent_tidptr = SCARG(uap, parent_tidptr);
tls = SCARG(uap, tls);
child_tidptr = SCARG(uap, child_tidptr);
tnprocs = atomic_inc_uint_nv(&nprocs);
if (__predict_false(tnprocs >= maxproc) ||
kauth_authorize_process(l->l_cred, KAUTH_PROCESS_FORK, p,
KAUTH_ARG(tnprocs), NULL, NULL) != 0) {
atomic_dec_uint(&nprocs);
return EAGAIN;
}
uaddr = uvm_uarea_alloc();
if (__predict_false(uaddr == 0)) {
atomic_dec_uint(&nprocs);
return ENOMEM;
}
error = lwp_create(l, p, uaddr, LWP_DETACHED | LWP_PIDLID,
SCARG(uap, stack), 0, child_return, NULL, &l2, l->l_class);
if (__predict_false(error)) {
DPRINTF(("%s: lwp_create error=%d\n", __func__, error));
atomic_dec_uint(&nprocs);
uvm_uarea_free(uaddr);
return error;
}
lid = l2->l_lid;
/* LINUX_CLONE_CHILD_CLEARTID: clear TID in child's memory on exit() */
if (flags & LINUX_CLONE_CHILD_CLEARTID) {
led = l2->l_emuldata;
led->led_clear_tid = child_tidptr;
}
/* LINUX_CLONE_PARENT_SETTID: store child's TID in parent's memory */
if (flags & LINUX_CLONE_PARENT_SETTID) {
if ((error = copyout(&lid, parent_tidptr, sizeof(lid))) != 0)
printf("%s: LINUX_CLONE_PARENT_SETTID "
"failed (parent_tidptr = %p tid = %d error=%d)\n",
__func__, parent_tidptr, lid, error);
}
/* LINUX_CLONE_CHILD_SETTID: store child's TID in child's memory */
if (flags & LINUX_CLONE_CHILD_SETTID) {
if ((error = copyout(&lid, child_tidptr, sizeof(lid))) != 0)
printf("%s: LINUX_CLONE_CHILD_SETTID "
"failed (child_tidptr = %p, tid = %d error=%d)\n",
__func__, child_tidptr, lid, error);
}
if (flags & LINUX_CLONE_SETTLS) {
error = LINUX_LWP_SETPRIVATE(l2, tls);
if (error) {
DPRINTF(("%s: LINUX_LWP_SETPRIVATE %d\n", __func__,
error));
lwp_exit(l2);
return error;
}
}
/*
* Set the new LWP running, unless the process is stopping,
* then the LWP is created stopped.
*/
mutex_enter(p->p_lock);
lwp_lock(l2);
spc = &l2->l_cpu->ci_schedstate;
if ((l->l_flag & (LW_WREBOOT | LW_WSUSPEND | LW_WEXIT)) == 0) {
if (p->p_stat == SSTOP || (p->p_sflag & PS_STOPPING) != 0) {
KASSERT(l2->l_wchan == NULL);
l2->l_stat = LSSTOP;
p->p_nrlwps--;
lwp_unlock_to(l2, spc->spc_lwplock);
} else {
KASSERT(lwp_locked(l2, spc->spc_mutex));
l2->l_stat = LSRUN;
sched_enqueue(l2, false);
lwp_unlock(l2);
}
} else {
l2->l_stat = LSSUSPENDED;
p->p_nrlwps--;
lwp_unlock_to(l2, spc->spc_lwplock);
}
mutex_exit(p->p_lock);
retval[0] = lid;
retval[1] = 0;
return 0;
}
